html_url
stringlengths 48
51
| title
stringlengths 5
280
| comments
stringlengths 63
51.8k
| body
stringlengths 0
36.2k
⌀ | comment_length
int64 16
1.52k
| text
stringlengths 163
54.1k
|
|---|---|---|---|---|---|
https://github.com/huggingface/datasets/issues/373 | Segmentation fault when loading local JSON dataset as of #372 | This seems like a more sensible solution! Thanks, @thomwolf. It's been a little daunting to understand what these scripts actually do, due to the level of abstraction and central documentation.
Am I correct in assuming that the `_generate_examples()` function is the actual procedure for how the data is loaded from file? Meaning that essentially with a file containing another format, that is the only function that requires re-implementation? I'm working with a lot of datasets that, due to licensing and privacy, cannot be published. As this library is so neatly integrated with the transformers library and gives easy access to public sets such as SQUAD and increased performance, it is very neat to be able to load my private sets as well. As of now, I have just been working on scripts for translating all my data into the SQUAD-format before using the json script, but I see that it might not be necessary after all. | The last issue was closed (#369) once the #372 update was merged. However, I'm still not able to load a SQuAD formatted JSON file. Instead of the previously recorded pyarrow error, I now get a segmentation fault.
```
dataset = nlp.load_dataset('json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
```
causes
```
Using custom data configuration default
Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, total: Unknown size) to /home/XXX/.cache/huggingface/datasets/json/default/0.0.0...
0 tables [00:00, ? tables/s]Segmentation fault (core dumped)
```
where `./datasets/train-v2.0.json` is downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/.
This is consistent with other SQuAD-formatted JSON files.
When attempting to load the dataset again, I get the following:
```
Using custom data configuration default
Traceback (most recent call last):
File "dataloader.py", line 6, in <module>
'json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 382, in download_and_prepare
with incomplete_dir(self._cache_dir) as tmp_data_dir:
File "/home/XXX/.conda/envs/torch/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 368, in incomplete_dir
os.makedirs(tmp_dir)
File "/home/XXX/.conda/envs/torch/lib/python3.7/os.py", line 223, in makedirs
mkdir(name, mode)
FileExistsError: [Errno 17] File exists: '/home/XXX/.cache/huggingface/datasets/json/default/0.0.0.incomplete'
```
(Not sure if you wanted this in the previous issue #369 or not as it was closed.) | 156 | Segmentation fault when loading local JSON dataset as of #372
The last issue was closed (#369) once the #372 update was merged. However, I'm still not able to load a SQuAD formatted JSON file. Instead of the previously recorded pyarrow error, I now get a segmentation fault.
```
dataset = nlp.load_dataset('json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
```
causes
```
Using custom data configuration default
Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, total: Unknown size) to /home/XXX/.cache/huggingface/datasets/json/default/0.0.0...
0 tables [00:00, ? tables/s]Segmentation fault (core dumped)
```
where `./datasets/train-v2.0.json` is downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/.
This is consistent with other SQuAD-formatted JSON files.
When attempting to load the dataset again, I get the following:
```
Using custom data configuration default
Traceback (most recent call last):
File "dataloader.py", line 6, in <module>
'json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 382, in download_and_prepare
with incomplete_dir(self._cache_dir) as tmp_data_dir:
File "/home/XXX/.conda/envs/torch/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 368, in incomplete_dir
os.makedirs(tmp_dir)
File "/home/XXX/.conda/envs/torch/lib/python3.7/os.py", line 223, in makedirs
mkdir(name, mode)
FileExistsError: [Errno 17] File exists: '/home/XXX/.cache/huggingface/datasets/json/default/0.0.0.incomplete'
```
(Not sure if you wanted this in the previous issue #369 or not as it was closed.)
This seems like a more sensible solution! Thanks, @thomwolf. It's been a little daunting to understand what these scripts actually do, due to the level of abstraction and central documentation.
Am I correct in assuming that the `_generate_examples()` function is the actual procedure for how the data is loaded from file? Meaning that essentially with a file containing another format, that is the only function that requires re-implementation? I'm working with a lot of datasets that, due to licensing and privacy, cannot be published. As this library is so neatly integrated with the transformers library and gives easy access to public sets such as SQUAD and increased performance, it is very neat to be able to load my private sets as well. As of now, I have just been working on scripts for translating all my data into the SQUAD-format before using the json script, but I see that it might not be necessary after all. |
https://github.com/huggingface/datasets/issues/373 | Segmentation fault when loading local JSON dataset as of #372 | Yes `_generate_examples()` is the main entry point. If you change the shape of the returned dictionary you also need to update the `features` in the `_info`.
I'm currently writing the doc so it should be easier soon to use the library and know how to add your datasets.
| The last issue was closed (#369) once the #372 update was merged. However, I'm still not able to load a SQuAD formatted JSON file. Instead of the previously recorded pyarrow error, I now get a segmentation fault.
```
dataset = nlp.load_dataset('json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
```
causes
```
Using custom data configuration default
Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, total: Unknown size) to /home/XXX/.cache/huggingface/datasets/json/default/0.0.0...
0 tables [00:00, ? tables/s]Segmentation fault (core dumped)
```
where `./datasets/train-v2.0.json` is downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/.
This is consistent with other SQuAD-formatted JSON files.
When attempting to load the dataset again, I get the following:
```
Using custom data configuration default
Traceback (most recent call last):
File "dataloader.py", line 6, in <module>
'json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 382, in download_and_prepare
with incomplete_dir(self._cache_dir) as tmp_data_dir:
File "/home/XXX/.conda/envs/torch/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 368, in incomplete_dir
os.makedirs(tmp_dir)
File "/home/XXX/.conda/envs/torch/lib/python3.7/os.py", line 223, in makedirs
mkdir(name, mode)
FileExistsError: [Errno 17] File exists: '/home/XXX/.cache/huggingface/datasets/json/default/0.0.0.incomplete'
```
(Not sure if you wanted this in the previous issue #369 or not as it was closed.) | 48 | Segmentation fault when loading local JSON dataset as of #372
The last issue was closed (#369) once the #372 update was merged. However, I'm still not able to load a SQuAD formatted JSON file. Instead of the previously recorded pyarrow error, I now get a segmentation fault.
```
dataset = nlp.load_dataset('json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
```
causes
```
Using custom data configuration default
Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, total: Unknown size) to /home/XXX/.cache/huggingface/datasets/json/default/0.0.0...
0 tables [00:00, ? tables/s]Segmentation fault (core dumped)
```
where `./datasets/train-v2.0.json` is downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/.
This is consistent with other SQuAD-formatted JSON files.
When attempting to load the dataset again, I get the following:
```
Using custom data configuration default
Traceback (most recent call last):
File "dataloader.py", line 6, in <module>
'json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 382, in download_and_prepare
with incomplete_dir(self._cache_dir) as tmp_data_dir:
File "/home/XXX/.conda/envs/torch/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 368, in incomplete_dir
os.makedirs(tmp_dir)
File "/home/XXX/.conda/envs/torch/lib/python3.7/os.py", line 223, in makedirs
mkdir(name, mode)
FileExistsError: [Errno 17] File exists: '/home/XXX/.cache/huggingface/datasets/json/default/0.0.0.incomplete'
```
(Not sure if you wanted this in the previous issue #369 or not as it was closed.)
Yes `_generate_examples()` is the main entry point. If you change the shape of the returned dictionary you also need to update the `features` in the `_info`.
I'm currently writing the doc so it should be easier soon to use the library and know how to add your datasets.
|
https://github.com/huggingface/datasets/issues/373 | Segmentation fault when loading local JSON dataset as of #372 | Could you try to update pyarrow to >=0.17.0 @vegarab ?
I don't have any segmentation fault with my version of pyarrow (0.17.1)
I tested with
```python
import nlp
s = nlp.load_dataset("json", data_files="train-v2.0.json", field="data", split="train")
s[0]
# {'title': 'Normans', 'paragraphs': [{'qas': [{'question': 'In what country is Normandy located?', 'id':...
``` | The last issue was closed (#369) once the #372 update was merged. However, I'm still not able to load a SQuAD formatted JSON file. Instead of the previously recorded pyarrow error, I now get a segmentation fault.
```
dataset = nlp.load_dataset('json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
```
causes
```
Using custom data configuration default
Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, total: Unknown size) to /home/XXX/.cache/huggingface/datasets/json/default/0.0.0...
0 tables [00:00, ? tables/s]Segmentation fault (core dumped)
```
where `./datasets/train-v2.0.json` is downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/.
This is consistent with other SQuAD-formatted JSON files.
When attempting to load the dataset again, I get the following:
```
Using custom data configuration default
Traceback (most recent call last):
File "dataloader.py", line 6, in <module>
'json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 382, in download_and_prepare
with incomplete_dir(self._cache_dir) as tmp_data_dir:
File "/home/XXX/.conda/envs/torch/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 368, in incomplete_dir
os.makedirs(tmp_dir)
File "/home/XXX/.conda/envs/torch/lib/python3.7/os.py", line 223, in makedirs
mkdir(name, mode)
FileExistsError: [Errno 17] File exists: '/home/XXX/.cache/huggingface/datasets/json/default/0.0.0.incomplete'
```
(Not sure if you wanted this in the previous issue #369 or not as it was closed.) | 49 | Segmentation fault when loading local JSON dataset as of #372
The last issue was closed (#369) once the #372 update was merged. However, I'm still not able to load a SQuAD formatted JSON file. Instead of the previously recorded pyarrow error, I now get a segmentation fault.
```
dataset = nlp.load_dataset('json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
```
causes
```
Using custom data configuration default
Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, total: Unknown size) to /home/XXX/.cache/huggingface/datasets/json/default/0.0.0...
0 tables [00:00, ? tables/s]Segmentation fault (core dumped)
```
where `./datasets/train-v2.0.json` is downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/.
This is consistent with other SQuAD-formatted JSON files.
When attempting to load the dataset again, I get the following:
```
Using custom data configuration default
Traceback (most recent call last):
File "dataloader.py", line 6, in <module>
'json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 382, in download_and_prepare
with incomplete_dir(self._cache_dir) as tmp_data_dir:
File "/home/XXX/.conda/envs/torch/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 368, in incomplete_dir
os.makedirs(tmp_dir)
File "/home/XXX/.conda/envs/torch/lib/python3.7/os.py", line 223, in makedirs
mkdir(name, mode)
FileExistsError: [Errno 17] File exists: '/home/XXX/.cache/huggingface/datasets/json/default/0.0.0.incomplete'
```
(Not sure if you wanted this in the previous issue #369 or not as it was closed.)
Could you try to update pyarrow to >=0.17.0 @vegarab ?
I don't have any segmentation fault with my version of pyarrow (0.17.1)
I tested with
```python
import nlp
s = nlp.load_dataset("json", data_files="train-v2.0.json", field="data", split="train")
s[0]
# {'title': 'Normans', 'paragraphs': [{'qas': [{'question': 'In what country is Normandy located?', 'id':...
``` |
https://github.com/huggingface/datasets/issues/373 | Segmentation fault when loading local JSON dataset as of #372 | Also if you want to have your own dataset script, we now have a new documentation !
See here:
https://huggingface.co/nlp/add_dataset.html | The last issue was closed (#369) once the #372 update was merged. However, I'm still not able to load a SQuAD formatted JSON file. Instead of the previously recorded pyarrow error, I now get a segmentation fault.
```
dataset = nlp.load_dataset('json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
```
causes
```
Using custom data configuration default
Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, total: Unknown size) to /home/XXX/.cache/huggingface/datasets/json/default/0.0.0...
0 tables [00:00, ? tables/s]Segmentation fault (core dumped)
```
where `./datasets/train-v2.0.json` is downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/.
This is consistent with other SQuAD-formatted JSON files.
When attempting to load the dataset again, I get the following:
```
Using custom data configuration default
Traceback (most recent call last):
File "dataloader.py", line 6, in <module>
'json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 382, in download_and_prepare
with incomplete_dir(self._cache_dir) as tmp_data_dir:
File "/home/XXX/.conda/envs/torch/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 368, in incomplete_dir
os.makedirs(tmp_dir)
File "/home/XXX/.conda/envs/torch/lib/python3.7/os.py", line 223, in makedirs
mkdir(name, mode)
FileExistsError: [Errno 17] File exists: '/home/XXX/.cache/huggingface/datasets/json/default/0.0.0.incomplete'
```
(Not sure if you wanted this in the previous issue #369 or not as it was closed.) | 20 | Segmentation fault when loading local JSON dataset as of #372
The last issue was closed (#369) once the #372 update was merged. However, I'm still not able to load a SQuAD formatted JSON file. Instead of the previously recorded pyarrow error, I now get a segmentation fault.
```
dataset = nlp.load_dataset('json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
```
causes
```
Using custom data configuration default
Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, total: Unknown size) to /home/XXX/.cache/huggingface/datasets/json/default/0.0.0...
0 tables [00:00, ? tables/s]Segmentation fault (core dumped)
```
where `./datasets/train-v2.0.json` is downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/.
This is consistent with other SQuAD-formatted JSON files.
When attempting to load the dataset again, I get the following:
```
Using custom data configuration default
Traceback (most recent call last):
File "dataloader.py", line 6, in <module>
'json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 382, in download_and_prepare
with incomplete_dir(self._cache_dir) as tmp_data_dir:
File "/home/XXX/.conda/envs/torch/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 368, in incomplete_dir
os.makedirs(tmp_dir)
File "/home/XXX/.conda/envs/torch/lib/python3.7/os.py", line 223, in makedirs
mkdir(name, mode)
FileExistsError: [Errno 17] File exists: '/home/XXX/.cache/huggingface/datasets/json/default/0.0.0.incomplete'
```
(Not sure if you wanted this in the previous issue #369 or not as it was closed.)
Also if you want to have your own dataset script, we now have a new documentation !
See here:
https://huggingface.co/nlp/add_dataset.html |
https://github.com/huggingface/datasets/issues/373 | Segmentation fault when loading local JSON dataset as of #372 | @lhoestq
For some reason, I am not able to reproduce the segmentation fault, on pyarrow==0.16.0. Using the exact same environment and file.
Anyhow, I discovered that pyarrow>=0.17.0 is required to read in a JSON file where the pandas structs contain lists. Otherwise, pyarrow complains when attempting to cast the struct:
```py
import nlp
>>> s = nlp.load_dataset("json", data_files="datasets/train-v2.0.json", field="data", split="train")
Using custom data configuration default
>>> s[0]
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/vegarab/.conda/envs/torch/lib/python3.7/site-packages/nlp/arrow_dataset.py", line 558, in __getitem__
format_kwargs=self._format_kwargs,
File "/home/vegarab/.conda/envs/torch/lib/python3.7/site-packages/nlp/arrow_dataset.py", line 498, in _getitem
outputs = self._unnest(self._data.slice(key, 1).to_pandas().to_dict("list"))
File "pyarrow/array.pxi", line 559, in pyarrow.lib._PandasConvertible.to_pandas
File "pyarrow/table.pxi", line 1367, in pyarrow.lib.Table._to_pandas
File "/home/vegarab/.conda/envs/torch/lib/python3.7/site-packages/pyarrow/pandas_compat.py", line 766, in table_to_blockmanager
blocks = _table_to_blocks(options, table, categories, ext_columns_dtypes)
File "/home/vegarab/.conda/envs/torch/lib/python3.7/site-packages/pyarrow/pandas_compat.py", line 1101, in _table_to_blocks
list(extension_columns.keys()))
File "pyarrow/table.pxi", line 881, in pyarrow.lib.table_to_blocks
File "pyarrow/error.pxi", line 105, in pyarrow.lib.check_status
pyarrow.lib.ArrowNotImplementedError: Not implemented type for Arrow list to pandas: struct<qas: list<item: struct<question: string, id: string, answers: list<item: struct<text: string, answer_start: int64>>, is_impossible: bool, plausible_answers: list<item: struct<text: string, answer_start: int64>>>>, context: string>
>>> s
Dataset(schema: {'title': 'string', 'paragraphs': 'list<item: struct<qas: list<item: struct<question: string, id: string, answers: list<item: struct<text: string, answer_start: int64>>, is_impossible: bool, plausible_answers: list<item: struct<text: string, answer_start: int64>>>>, context: string>>'}, num_rows: 35)
```
Upgrading to >=0.17.0 provides the same dataset structure, but accessing the records is possible without the same exception.
| The last issue was closed (#369) once the #372 update was merged. However, I'm still not able to load a SQuAD formatted JSON file. Instead of the previously recorded pyarrow error, I now get a segmentation fault.
```
dataset = nlp.load_dataset('json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
```
causes
```
Using custom data configuration default
Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, total: Unknown size) to /home/XXX/.cache/huggingface/datasets/json/default/0.0.0...
0 tables [00:00, ? tables/s]Segmentation fault (core dumped)
```
where `./datasets/train-v2.0.json` is downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/.
This is consistent with other SQuAD-formatted JSON files.
When attempting to load the dataset again, I get the following:
```
Using custom data configuration default
Traceback (most recent call last):
File "dataloader.py", line 6, in <module>
'json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 382, in download_and_prepare
with incomplete_dir(self._cache_dir) as tmp_data_dir:
File "/home/XXX/.conda/envs/torch/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 368, in incomplete_dir
os.makedirs(tmp_dir)
File "/home/XXX/.conda/envs/torch/lib/python3.7/os.py", line 223, in makedirs
mkdir(name, mode)
FileExistsError: [Errno 17] File exists: '/home/XXX/.cache/huggingface/datasets/json/default/0.0.0.incomplete'
```
(Not sure if you wanted this in the previous issue #369 or not as it was closed.) | 219 | Segmentation fault when loading local JSON dataset as of #372
The last issue was closed (#369) once the #372 update was merged. However, I'm still not able to load a SQuAD formatted JSON file. Instead of the previously recorded pyarrow error, I now get a segmentation fault.
```
dataset = nlp.load_dataset('json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
```
causes
```
Using custom data configuration default
Downloading and preparing dataset json/default (download: Unknown size, generated: Unknown size, total: Unknown size) to /home/XXX/.cache/huggingface/datasets/json/default/0.0.0...
0 tables [00:00, ? tables/s]Segmentation fault (core dumped)
```
where `./datasets/train-v2.0.json` is downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/.
This is consistent with other SQuAD-formatted JSON files.
When attempting to load the dataset again, I get the following:
```
Using custom data configuration default
Traceback (most recent call last):
File "dataloader.py", line 6, in <module>
'json', data_files={nlp.Split.TRAIN: ["./datasets/train-v2.0.json"]}, field='data')
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 382, in download_and_prepare
with incomplete_dir(self._cache_dir) as tmp_data_dir:
File "/home/XXX/.conda/envs/torch/lib/python3.7/contextlib.py", line 112, in __enter__
return next(self.gen)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 368, in incomplete_dir
os.makedirs(tmp_dir)
File "/home/XXX/.conda/envs/torch/lib/python3.7/os.py", line 223, in makedirs
mkdir(name, mode)
FileExistsError: [Errno 17] File exists: '/home/XXX/.cache/huggingface/datasets/json/default/0.0.0.incomplete'
```
(Not sure if you wanted this in the previous issue #369 or not as it was closed.)
@lhoestq
For some reason, I am not able to reproduce the segmentation fault, on pyarrow==0.16.0. Using the exact same environment and file.
Anyhow, I discovered that pyarrow>=0.17.0 is required to read in a JSON file where the pandas structs contain lists. Otherwise, pyarrow complains when attempting to cast the struct:
```py
import nlp
>>> s = nlp.load_dataset("json", data_files="datasets/train-v2.0.json", field="data", split="train")
Using custom data configuration default
>>> s[0]
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/vegarab/.conda/envs/torch/lib/python3.7/site-packages/nlp/arrow_dataset.py", line 558, in __getitem__
format_kwargs=self._format_kwargs,
File "/home/vegarab/.conda/envs/torch/lib/python3.7/site-packages/nlp/arrow_dataset.py", line 498, in _getitem
outputs = self._unnest(self._data.slice(key, 1).to_pandas().to_dict("list"))
File "pyarrow/array.pxi", line 559, in pyarrow.lib._PandasConvertible.to_pandas
File "pyarrow/table.pxi", line 1367, in pyarrow.lib.Table._to_pandas
File "/home/vegarab/.conda/envs/torch/lib/python3.7/site-packages/pyarrow/pandas_compat.py", line 766, in table_to_blockmanager
blocks = _table_to_blocks(options, table, categories, ext_columns_dtypes)
File "/home/vegarab/.conda/envs/torch/lib/python3.7/site-packages/pyarrow/pandas_compat.py", line 1101, in _table_to_blocks
list(extension_columns.keys()))
File "pyarrow/table.pxi", line 881, in pyarrow.lib.table_to_blocks
File "pyarrow/error.pxi", line 105, in pyarrow.lib.check_status
pyarrow.lib.ArrowNotImplementedError: Not implemented type for Arrow list to pandas: struct<qas: list<item: struct<question: string, id: string, answers: list<item: struct<text: string, answer_start: int64>>, is_impossible: bool, plausible_answers: list<item: struct<text: string, answer_start: int64>>>>, context: string>
>>> s
Dataset(schema: {'title': 'string', 'paragraphs': 'list<item: struct<qas: list<item: struct<question: string, id: string, answers: list<item: struct<text: string, answer_start: int64>>, is_impossible: bool, plausible_answers: list<item: struct<text: string, answer_start: int64>>>>, context: string>>'}, num_rows: 35)
```
Upgrading to >=0.17.0 provides the same dataset structure, but accessing the records is possible without the same exception.
|
https://github.com/huggingface/datasets/issues/369 | can't load local dataset: pyarrow.lib.ArrowInvalid: straddling object straddles two block boundaries | I am able to reproduce this with the official SQuAD `train-v2.0.json` file downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/ | Trying to load a local SQuAD-formatted dataset (from a JSON file, about 60MB):
```
dataset = nlp.load_dataset(path='json', data_files={nlp.Split.TRAIN: ["./path/to/file.json"]})
```
causes
```
Traceback (most recent call last):
File "dataloader.py", line 9, in <module>
["./path/to/file.json"]})
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 432, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 483, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 719, in _prepare_split
for key, table in utils.tqdm(generator, unit=" tables", leave=False):
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/tqdm/std.py", line 1129, in __iter__
for obj in iterable:
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/datasets/json/88c1bc5c68489f7eda549ed05a5a738527c613b3e7a4ee3524d9d233353a949b/json.py", line 53, in _generate_tables
file, read_options=self.config.pa_read_options, parse_options=self.config.pa_parse_options,
File "pyarrow/_json.pyx", line 191, in pyarrow._json.read_json
File "pyarrow/error.pxi", line 85, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: straddling object straddles two block boundaries (try to increase block size?)
```
I haven't been able to find any reports of this specific pyarrow error here or elsewhere. | 16 | can't load local dataset: pyarrow.lib.ArrowInvalid: straddling object straddles two block boundaries
Trying to load a local SQuAD-formatted dataset (from a JSON file, about 60MB):
```
dataset = nlp.load_dataset(path='json', data_files={nlp.Split.TRAIN: ["./path/to/file.json"]})
```
causes
```
Traceback (most recent call last):
File "dataloader.py", line 9, in <module>
["./path/to/file.json"]})
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 432, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 483, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 719, in _prepare_split
for key, table in utils.tqdm(generator, unit=" tables", leave=False):
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/tqdm/std.py", line 1129, in __iter__
for obj in iterable:
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/datasets/json/88c1bc5c68489f7eda549ed05a5a738527c613b3e7a4ee3524d9d233353a949b/json.py", line 53, in _generate_tables
file, read_options=self.config.pa_read_options, parse_options=self.config.pa_parse_options,
File "pyarrow/_json.pyx", line 191, in pyarrow._json.read_json
File "pyarrow/error.pxi", line 85, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: straddling object straddles two block boundaries (try to increase block size?)
```
I haven't been able to find any reports of this specific pyarrow error here or elsewhere.
I am able to reproduce this with the official SQuAD `train-v2.0.json` file downloaded directly from https://rajpurkar.github.io/SQuAD-explorer/ |
https://github.com/huggingface/datasets/issues/369 | can't load local dataset: pyarrow.lib.ArrowInvalid: straddling object straddles two block boundaries | I am facing this issue in transformers library 3.0.2 while reading a csv using datasets.
Is this fixed in latest version?
I updated the latest version 4.0.1 but still getting this error. What could cause this error? | Trying to load a local SQuAD-formatted dataset (from a JSON file, about 60MB):
```
dataset = nlp.load_dataset(path='json', data_files={nlp.Split.TRAIN: ["./path/to/file.json"]})
```
causes
```
Traceback (most recent call last):
File "dataloader.py", line 9, in <module>
["./path/to/file.json"]})
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 432, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 483, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 719, in _prepare_split
for key, table in utils.tqdm(generator, unit=" tables", leave=False):
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/tqdm/std.py", line 1129, in __iter__
for obj in iterable:
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/datasets/json/88c1bc5c68489f7eda549ed05a5a738527c613b3e7a4ee3524d9d233353a949b/json.py", line 53, in _generate_tables
file, read_options=self.config.pa_read_options, parse_options=self.config.pa_parse_options,
File "pyarrow/_json.pyx", line 191, in pyarrow._json.read_json
File "pyarrow/error.pxi", line 85, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: straddling object straddles two block boundaries (try to increase block size?)
```
I haven't been able to find any reports of this specific pyarrow error here or elsewhere. | 37 | can't load local dataset: pyarrow.lib.ArrowInvalid: straddling object straddles two block boundaries
Trying to load a local SQuAD-formatted dataset (from a JSON file, about 60MB):
```
dataset = nlp.load_dataset(path='json', data_files={nlp.Split.TRAIN: ["./path/to/file.json"]})
```
causes
```
Traceback (most recent call last):
File "dataloader.py", line 9, in <module>
["./path/to/file.json"]})
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/load.py", line 524, in load_dataset
save_infos=save_infos,
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 432, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 483, in _download_and_prepare
self._prepare_split(split_generator, **prepare_split_kwargs)
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/builder.py", line 719, in _prepare_split
for key, table in utils.tqdm(generator, unit=" tables", leave=False):
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/tqdm/std.py", line 1129, in __iter__
for obj in iterable:
File "/home/XXX/.conda/envs/torch/lib/python3.7/site-packages/nlp/datasets/json/88c1bc5c68489f7eda549ed05a5a738527c613b3e7a4ee3524d9d233353a949b/json.py", line 53, in _generate_tables
file, read_options=self.config.pa_read_options, parse_options=self.config.pa_parse_options,
File "pyarrow/_json.pyx", line 191, in pyarrow._json.read_json
File "pyarrow/error.pxi", line 85, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: straddling object straddles two block boundaries (try to increase block size?)
```
I haven't been able to find any reports of this specific pyarrow error here or elsewhere.
I am facing this issue in transformers library 3.0.2 while reading a csv using datasets.
Is this fixed in latest version?
I updated the latest version 4.0.1 but still getting this error. What could cause this error? |
https://github.com/huggingface/datasets/issues/368 | load_metric can't acquire lock anymore | I found that, in the same process (or the same interactive session), if I do
import nlp
m1 = nlp.load_metric('glue', 'mrpc')
m2 = nlp.load_metric('glue', 'sst2')
I will get the same error `ValueError: Cannot acquire lock, caching file might be used by another process, you should setup a unique 'experiment_id'`. | I can't load metric (glue) anymore after an error in a previous run. I even removed the whole cache folder `/home/XXX/.cache/huggingface/`, and the issue persisted. What are the steps to fix this?
Traceback (most recent call last):
File "/home/XXX/miniconda3/envs/ML-DL-py-3.7/lib/python3.7/site-packages/nlp/metric.py", line 101, in __init__
self.filelock.acquire(timeout=1)
File "/home/XXX/miniconda3/envs/ML-DL-py-3.7/lib/python3.7/site-packages/filelock.py", line 278, in acquire
raise Timeout(self._lock_file)
filelock.Timeout: The file lock '/home/XXX/.cache/huggingface/metrics/glue/1.0.0/1-glue-0.arrow.lock' could not be acquired.
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "examples_huggingface_nlp.py", line 268, in <module>
main()
File "examples_huggingface_nlp.py", line 242, in main
dataset, metric = get_dataset_metric(glue_task)
File "examples_huggingface_nlp.py", line 77, in get_dataset_metric
metric = nlp.load_metric('glue', glue_config, experiment_id=1)
File "/home/XXX/miniconda3/envs/ML-DL-py-3.7/lib/python3.7/site-packages/nlp/load.py", line 440, in load_metric
**metric_init_kwargs,
File "/home/XXX/miniconda3/envs/ML-DL-py-3.7/lib/python3.7/site-packages/nlp/metric.py", line 104, in __init__
"Cannot acquire lock, caching file might be used by another process, "
ValueError: Cannot acquire lock, caching file might be used by another process, you should setup a unique 'experiment_id' for this run.
I0709 15:54:41.008838 139854118430464 filelock.py:318] Lock 139852058030936 released on /home/XXX/.cache/huggingface/metrics/glue/1.0.0/1-glue-0.arrow.lock
| 49 | load_metric can't acquire lock anymore
I can't load metric (glue) anymore after an error in a previous run. I even removed the whole cache folder `/home/XXX/.cache/huggingface/`, and the issue persisted. What are the steps to fix this?
Traceback (most recent call last):
File "/home/XXX/miniconda3/envs/ML-DL-py-3.7/lib/python3.7/site-packages/nlp/metric.py", line 101, in __init__
self.filelock.acquire(timeout=1)
File "/home/XXX/miniconda3/envs/ML-DL-py-3.7/lib/python3.7/site-packages/filelock.py", line 278, in acquire
raise Timeout(self._lock_file)
filelock.Timeout: The file lock '/home/XXX/.cache/huggingface/metrics/glue/1.0.0/1-glue-0.arrow.lock' could not be acquired.
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "examples_huggingface_nlp.py", line 268, in <module>
main()
File "examples_huggingface_nlp.py", line 242, in main
dataset, metric = get_dataset_metric(glue_task)
File "examples_huggingface_nlp.py", line 77, in get_dataset_metric
metric = nlp.load_metric('glue', glue_config, experiment_id=1)
File "/home/XXX/miniconda3/envs/ML-DL-py-3.7/lib/python3.7/site-packages/nlp/load.py", line 440, in load_metric
**metric_init_kwargs,
File "/home/XXX/miniconda3/envs/ML-DL-py-3.7/lib/python3.7/site-packages/nlp/metric.py", line 104, in __init__
"Cannot acquire lock, caching file might be used by another process, "
ValueError: Cannot acquire lock, caching file might be used by another process, you should setup a unique 'experiment_id' for this run.
I0709 15:54:41.008838 139854118430464 filelock.py:318] Lock 139852058030936 released on /home/XXX/.cache/huggingface/metrics/glue/1.0.0/1-glue-0.arrow.lock
I found that, in the same process (or the same interactive session), if I do
import nlp
m1 = nlp.load_metric('glue', 'mrpc')
m2 = nlp.load_metric('glue', 'sst2')
I will get the same error `ValueError: Cannot acquire lock, caching file might be used by another process, you should setup a unique 'experiment_id'`. |
https://github.com/huggingface/datasets/issues/365 | How to augment data ? | Using batched map is probably the easiest way at the moment.
What kind of augmentation would you like to do ? | Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
``` | 21 | How to augment data ?
Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
```
Using batched map is probably the easiest way at the moment.
What kind of augmentation would you like to do ? |
https://github.com/huggingface/datasets/issues/365 | How to augment data ? | Some samples in the dataset are too long, I want to divide them in several samples. | Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
``` | 16 | How to augment data ?
Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
```
Some samples in the dataset are too long, I want to divide them in several samples. |
https://github.com/huggingface/datasets/issues/365 | How to augment data ? | Using batched map is the way to go then.
We'll make it clearer in the docs that map could be used for augmentation.
Let me know if you think there should be another way to do it. Or feel free to close the issue otherwise. | Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
``` | 45 | How to augment data ?
Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
```
Using batched map is the way to go then.
We'll make it clearer in the docs that map could be used for augmentation.
Let me know if you think there should be another way to do it. Or feel free to close the issue otherwise. |
https://github.com/huggingface/datasets/issues/365 | How to augment data ? | It just feels awkward to use map to augment data. Also it means it's not possible to augment data in a non-batched way.
But to be honest I have no idea of a good API... | Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
``` | 35 | How to augment data ?
Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
```
It just feels awkward to use map to augment data. Also it means it's not possible to augment data in a non-batched way.
But to be honest I have no idea of a good API... |
https://github.com/huggingface/datasets/issues/365 | How to augment data ? | Or for non-batched samples, how about returning a tuple ?
```python
def aug(sample):
# Simply copy the existing data to have x2 amount of data
return sample, sample
dataset = dataset.map(aug)
```
It feels really natural and easy, but :
* it means the behavior with batched data is different
* I don't know how doable it is backend-wise
@lhoestq | Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
``` | 60 | How to augment data ?
Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
```
Or for non-batched samples, how about returning a tuple ?
```python
def aug(sample):
# Simply copy the existing data to have x2 amount of data
return sample, sample
dataset = dataset.map(aug)
```
It feels really natural and easy, but :
* it means the behavior with batched data is different
* I don't know how doable it is backend-wise
@lhoestq |
https://github.com/huggingface/datasets/issues/365 | How to augment data ? | As we're working with arrow's columnar format we prefer to play with batches that are dictionaries instead of tuples.
If we have tuple it implies to re-format the data each time we want to write to arrow, which can lower the speed of map for example.
It's also a matter of coherence, as we don't want users to be confused whether they have to return dictionaries for some functions and tuples for others when they're doing batches. | Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
``` | 77 | How to augment data ?
Is there any clean way to augment data ?
For now my work-around is to use batched map, like this :
```python
def aug(samples):
# Simply copy the existing data to have x2 amount of data
for k, v in samples.items():
samples[k].extend(v)
return samples
dataset = dataset.map(aug, batched=True)
```
As we're working with arrow's columnar format we prefer to play with batches that are dictionaries instead of tuples.
If we have tuple it implies to re-format the data each time we want to write to arrow, which can lower the speed of map for example.
It's also a matter of coherence, as we don't want users to be confused whether they have to return dictionaries for some functions and tuples for others when they're doing batches. |
https://github.com/huggingface/datasets/issues/361 | 🐛 [Metrics] ROUGE is non-deterministic | > Hi, can you give a full self-contained example to reproduce this behavior?
There is a notebook in the post ;) | If I run the ROUGE metric 2 times, with same predictions / references, the scores are slightly different.
Refer to [this Colab notebook](https://colab.research.google.com/drive/1wRssNXgb9ldcp4ulwj-hMJn0ywhDOiDy?usp=sharing) for reproducing the problem.
Example of F-score for ROUGE-1, ROUGE-2, ROUGE-L in 2 differents run :
> ['0.3350', '0.1470', '0.2329']
['0.3358', '0.1451', '0.2332']
---
Why ROUGE is not deterministic ? | 21 | 🐛 [Metrics] ROUGE is non-deterministic
If I run the ROUGE metric 2 times, with same predictions / references, the scores are slightly different.
Refer to [this Colab notebook](https://colab.research.google.com/drive/1wRssNXgb9ldcp4ulwj-hMJn0ywhDOiDy?usp=sharing) for reproducing the problem.
Example of F-score for ROUGE-1, ROUGE-2, ROUGE-L in 2 differents run :
> ['0.3350', '0.1470', '0.2329']
['0.3358', '0.1451', '0.2332']
---
Why ROUGE is not deterministic ?
> Hi, can you give a full self-contained example to reproduce this behavior?
There is a notebook in the post ;) |
https://github.com/huggingface/datasets/issues/361 | 🐛 [Metrics] ROUGE is non-deterministic | > If I run the ROUGE metric 2 times, with same predictions / references, the scores are slightly different.
>
> Refer to [this Colab notebook](https://colab.research.google.com/drive/1wRssNXgb9ldcp4ulwj-hMJn0ywhDOiDy?usp=sharing) for reproducing the problem.
>
> Example of F-score for ROUGE-1, ROUGE-2, ROUGE-L in 2 differents run :
>
> > ['0.3350', '0.1470', '0.2329']
> > ['0.3358', '0.1451', '0.2332']
>
> Why ROUGE is not deterministic ?
This is because of rouge's `BootstrapAggregator` that uses sampling to get confidence intervals (low, mid, high).
You can get deterministic scores per sentence pair by using
```python
score = rouge.compute(rouge_types=["rouge1", "rouge2", "rougeL"], use_aggregator=False)
```
Or you can set numpy's random seed if you still want to use the aggregator. | If I run the ROUGE metric 2 times, with same predictions / references, the scores are slightly different.
Refer to [this Colab notebook](https://colab.research.google.com/drive/1wRssNXgb9ldcp4ulwj-hMJn0ywhDOiDy?usp=sharing) for reproducing the problem.
Example of F-score for ROUGE-1, ROUGE-2, ROUGE-L in 2 differents run :
> ['0.3350', '0.1470', '0.2329']
['0.3358', '0.1451', '0.2332']
---
Why ROUGE is not deterministic ? | 112 | 🐛 [Metrics] ROUGE is non-deterministic
If I run the ROUGE metric 2 times, with same predictions / references, the scores are slightly different.
Refer to [this Colab notebook](https://colab.research.google.com/drive/1wRssNXgb9ldcp4ulwj-hMJn0ywhDOiDy?usp=sharing) for reproducing the problem.
Example of F-score for ROUGE-1, ROUGE-2, ROUGE-L in 2 differents run :
> ['0.3350', '0.1470', '0.2329']
['0.3358', '0.1451', '0.2332']
---
Why ROUGE is not deterministic ?
> If I run the ROUGE metric 2 times, with same predictions / references, the scores are slightly different.
>
> Refer to [this Colab notebook](https://colab.research.google.com/drive/1wRssNXgb9ldcp4ulwj-hMJn0ywhDOiDy?usp=sharing) for reproducing the problem.
>
> Example of F-score for ROUGE-1, ROUGE-2, ROUGE-L in 2 differents run :
>
> > ['0.3350', '0.1470', '0.2329']
> > ['0.3358', '0.1451', '0.2332']
>
> Why ROUGE is not deterministic ?
This is because of rouge's `BootstrapAggregator` that uses sampling to get confidence intervals (low, mid, high).
You can get deterministic scores per sentence pair by using
```python
score = rouge.compute(rouge_types=["rouge1", "rouge2", "rougeL"], use_aggregator=False)
```
Or you can set numpy's random seed if you still want to use the aggregator. |
https://github.com/huggingface/datasets/issues/361 | 🐛 [Metrics] ROUGE is non-deterministic | Now if you re-run the notebook, the two printed results are the same @colanim
```
['0.3356', '0.1466', '0.2318']
['0.3356', '0.1466', '0.2318']
```
However across sessions, the results may change (as numpy's random seed can be different). You can prevent that by setting your seed:
```python
rouge = nlp.load_metric('rouge', seed=42)
``` | If I run the ROUGE metric 2 times, with same predictions / references, the scores are slightly different.
Refer to [this Colab notebook](https://colab.research.google.com/drive/1wRssNXgb9ldcp4ulwj-hMJn0ywhDOiDy?usp=sharing) for reproducing the problem.
Example of F-score for ROUGE-1, ROUGE-2, ROUGE-L in 2 differents run :
> ['0.3350', '0.1470', '0.2329']
['0.3358', '0.1451', '0.2332']
---
Why ROUGE is not deterministic ? | 50 | 🐛 [Metrics] ROUGE is non-deterministic
If I run the ROUGE metric 2 times, with same predictions / references, the scores are slightly different.
Refer to [this Colab notebook](https://colab.research.google.com/drive/1wRssNXgb9ldcp4ulwj-hMJn0ywhDOiDy?usp=sharing) for reproducing the problem.
Example of F-score for ROUGE-1, ROUGE-2, ROUGE-L in 2 differents run :
> ['0.3350', '0.1470', '0.2329']
['0.3358', '0.1451', '0.2332']
---
Why ROUGE is not deterministic ?
Now if you re-run the notebook, the two printed results are the same @colanim
```
['0.3356', '0.1466', '0.2318']
['0.3356', '0.1466', '0.2318']
```
However across sessions, the results may change (as numpy's random seed can be different). You can prevent that by setting your seed:
```python
rouge = nlp.load_metric('rouge', seed=42)
``` |
https://github.com/huggingface/datasets/issues/360 | [Feature request] Add dataset.ragged_map() function for many-to-many transformations | Actually `map(batched=True)` can already change the size of the dataset.
It can accept examples of length `N` and returns a batch of length `M` (can be null or greater than `N`).
I'll make that explicit in the doc that I'm currently writing. | `dataset.map()` enables one-to-one transformations. Input one example and output one example. This is helpful for tokenizing and cleaning individual lines.
`dataset.filter()` enables one-to-(one-or-none) transformations. Input one example and output either zero/one example. This is helpful for removing portions from the dataset.
However, some dataset transformations are many-to-many. Consider constructing BERT training examples from a dataset of sentences, where you map `["a", "b", "c"] -> ["a[SEP]b", "a[SEP]c", "b[SEP]c", "c[SEP]b", ...]`
I propose a more general `ragged_map()` method that takes in a batch of examples of length `N` and return a batch of examples `M`. This is different from the `map(batched=True)` method, which takes examples of length `N` and returns a batch of length `N`, processing individual examples in parallel. I don't have a clear vision of how this would be implemented efficiently and lazily, but would love to hear the community's feedback on this.
My specific use case is creating an end-to-end ELECTRA data pipeline. I would like to take the raw WikiText data and generate training examples from this using the `ragged_map()` method, then export to TFRecords and train quickly. This would be a reproducible pipeline with no bash scripts. Currently I'm relying on scripts like https://github.com/google-research/electra/blob/master/build_pretraining_dataset.py, which are less general.
| 42 | [Feature request] Add dataset.ragged_map() function for many-to-many transformations
`dataset.map()` enables one-to-one transformations. Input one example and output one example. This is helpful for tokenizing and cleaning individual lines.
`dataset.filter()` enables one-to-(one-or-none) transformations. Input one example and output either zero/one example. This is helpful for removing portions from the dataset.
However, some dataset transformations are many-to-many. Consider constructing BERT training examples from a dataset of sentences, where you map `["a", "b", "c"] -> ["a[SEP]b", "a[SEP]c", "b[SEP]c", "c[SEP]b", ...]`
I propose a more general `ragged_map()` method that takes in a batch of examples of length `N` and return a batch of examples `M`. This is different from the `map(batched=True)` method, which takes examples of length `N` and returns a batch of length `N`, processing individual examples in parallel. I don't have a clear vision of how this would be implemented efficiently and lazily, but would love to hear the community's feedback on this.
My specific use case is creating an end-to-end ELECTRA data pipeline. I would like to take the raw WikiText data and generate training examples from this using the `ragged_map()` method, then export to TFRecords and train quickly. This would be a reproducible pipeline with no bash scripts. Currently I'm relying on scripts like https://github.com/google-research/electra/blob/master/build_pretraining_dataset.py, which are less general.
Actually `map(batched=True)` can already change the size of the dataset.
It can accept examples of length `N` and returns a batch of length `M` (can be null or greater than `N`).
I'll make that explicit in the doc that I'm currently writing. |
https://github.com/huggingface/datasets/issues/360 | [Feature request] Add dataset.ragged_map() function for many-to-many transformations | You're two steps ahead of me :) In my testing, it also works if `M` < `N`.
A batched map of different length seems to work if you directly overwrite all of the original keys, but fails if any of the original keys are preserved.
For example,
```python
# Create a dummy dataset
dset = load_dataset("wikitext", "wikitext-2-raw-v1")["test"]
dset = dset.map(lambda ex: {"length": len(ex["text"]), "foo": 1})
# Do an allreduce on each batch, overwriting both keys
dset.map(lambda batch: {"length": [sum(batch["length"])], "foo": [1]})
# Dataset(schema: {'length': 'int64', 'foo': 'int64'}, num_rows: 5)
# Now attempt an allreduce without touching the `foo` key
dset.map(lambda batch: {"length": [sum(batch["length"])]})
# This fails with the error message below
```
```bash
File "/path/to/nlp/src/nlp/arrow_dataset.py", line 728, in map
arrow_schema = pa.Table.from_pydict(test_output).schema
File "pyarrow/io.pxi", line 1532, in pyarrow.lib.Codec.detect
File "pyarrow/table.pxi", line 1503, in pyarrow.lib.Table.from_arrays
File "pyarrow/public-api.pxi", line 390, in pyarrow.lib.pyarrow_wrap_table
File "pyarrow/error.pxi", line 85, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: Column 1 named foo expected length 1 but got length 2
```
Adding the `remove_columns=["length", "foo"]` argument to `map()` solves the issue. Leaving the above error for future visitors. Perfect, thank you! | `dataset.map()` enables one-to-one transformations. Input one example and output one example. This is helpful for tokenizing and cleaning individual lines.
`dataset.filter()` enables one-to-(one-or-none) transformations. Input one example and output either zero/one example. This is helpful for removing portions from the dataset.
However, some dataset transformations are many-to-many. Consider constructing BERT training examples from a dataset of sentences, where you map `["a", "b", "c"] -> ["a[SEP]b", "a[SEP]c", "b[SEP]c", "c[SEP]b", ...]`
I propose a more general `ragged_map()` method that takes in a batch of examples of length `N` and return a batch of examples `M`. This is different from the `map(batched=True)` method, which takes examples of length `N` and returns a batch of length `N`, processing individual examples in parallel. I don't have a clear vision of how this would be implemented efficiently and lazily, but would love to hear the community's feedback on this.
My specific use case is creating an end-to-end ELECTRA data pipeline. I would like to take the raw WikiText data and generate training examples from this using the `ragged_map()` method, then export to TFRecords and train quickly. This would be a reproducible pipeline with no bash scripts. Currently I'm relying on scripts like https://github.com/google-research/electra/blob/master/build_pretraining_dataset.py, which are less general.
| 179 | [Feature request] Add dataset.ragged_map() function for many-to-many transformations
`dataset.map()` enables one-to-one transformations. Input one example and output one example. This is helpful for tokenizing and cleaning individual lines.
`dataset.filter()` enables one-to-(one-or-none) transformations. Input one example and output either zero/one example. This is helpful for removing portions from the dataset.
However, some dataset transformations are many-to-many. Consider constructing BERT training examples from a dataset of sentences, where you map `["a", "b", "c"] -> ["a[SEP]b", "a[SEP]c", "b[SEP]c", "c[SEP]b", ...]`
I propose a more general `ragged_map()` method that takes in a batch of examples of length `N` and return a batch of examples `M`. This is different from the `map(batched=True)` method, which takes examples of length `N` and returns a batch of length `N`, processing individual examples in parallel. I don't have a clear vision of how this would be implemented efficiently and lazily, but would love to hear the community's feedback on this.
My specific use case is creating an end-to-end ELECTRA data pipeline. I would like to take the raw WikiText data and generate training examples from this using the `ragged_map()` method, then export to TFRecords and train quickly. This would be a reproducible pipeline with no bash scripts. Currently I'm relying on scripts like https://github.com/google-research/electra/blob/master/build_pretraining_dataset.py, which are less general.
You're two steps ahead of me :) In my testing, it also works if `M` < `N`.
A batched map of different length seems to work if you directly overwrite all of the original keys, but fails if any of the original keys are preserved.
For example,
```python
# Create a dummy dataset
dset = load_dataset("wikitext", "wikitext-2-raw-v1")["test"]
dset = dset.map(lambda ex: {"length": len(ex["text"]), "foo": 1})
# Do an allreduce on each batch, overwriting both keys
dset.map(lambda batch: {"length": [sum(batch["length"])], "foo": [1]})
# Dataset(schema: {'length': 'int64', 'foo': 'int64'}, num_rows: 5)
# Now attempt an allreduce without touching the `foo` key
dset.map(lambda batch: {"length": [sum(batch["length"])]})
# This fails with the error message below
```
```bash
File "/path/to/nlp/src/nlp/arrow_dataset.py", line 728, in map
arrow_schema = pa.Table.from_pydict(test_output).schema
File "pyarrow/io.pxi", line 1532, in pyarrow.lib.Codec.detect
File "pyarrow/table.pxi", line 1503, in pyarrow.lib.Table.from_arrays
File "pyarrow/public-api.pxi", line 390, in pyarrow.lib.pyarrow_wrap_table
File "pyarrow/error.pxi", line 85, in pyarrow.lib.check_status
pyarrow.lib.ArrowInvalid: Column 1 named foo expected length 1 but got length 2
```
Adding the `remove_columns=["length", "foo"]` argument to `map()` solves the issue. Leaving the above error for future visitors. Perfect, thank you! |
https://github.com/huggingface/datasets/issues/359 | ArrowBasedBuilder _prepare_split parse_schema breaks on nested structures | Hi, it depends on what it is in your `dataset_builder.py` file. Can you share it?
If you are just loading `json` files, you can also directly use the `json` script (which will find the schema/features from your JSON structure):
```python
from nlp import load_dataset
ds = load_dataset("json", data_files=rel_datafiles)
``` | I tried using the Json dataloader to load some JSON lines files. but get an exception in the parse_schema function.
```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-23-9aecfbee53bd> in <module>
55 from nlp import load_dataset
56
---> 57 ds = load_dataset("../text2struct/model/dataset_builder.py", data_files=rel_datafiles)
58
59
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
736 schema_dict[field.name] = Value(str(field.type))
737
--> 738 parse_schema(writer.schema, features)
739 self.info.features = Features(features)
740
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in parse_schema(schema, schema_dict)
734 parse_schema(field.type.value_type, schema_dict[field.name])
735 else:
--> 736 schema_dict[field.name] = Value(str(field.type))
737
738 parse_schema(writer.schema, features)
<string> in __init__(self, dtype, id, _type)
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in __post_init__(self)
55
56 def __post_init__(self):
---> 57 self.pa_type = string_to_arrow(self.dtype)
58
59 def __call__(self):
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in string_to_arrow(type_str)
32 if str(type_str + "_") not in pa.__dict__:
33 raise ValueError(
---> 34 f"Neither {type_str} nor {type_str + '_'} seems to be a pyarrow data type. "
35 f"Please make sure to use a correct data type, see: "
36 f"https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions"
ValueError: Neither list<item: string> nor list<item: string>_ seems to be a pyarrow data type. Please make sure to use a correct data type, see: https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions
```
If I create the dataset imperatively, using a pyarrow table, the dataset is created correctly. If I override the `_prepare_split` method to avoid calling the validate schema, the dataset can load as well. | 49 | ArrowBasedBuilder _prepare_split parse_schema breaks on nested structures
I tried using the Json dataloader to load some JSON lines files. but get an exception in the parse_schema function.
```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-23-9aecfbee53bd> in <module>
55 from nlp import load_dataset
56
---> 57 ds = load_dataset("../text2struct/model/dataset_builder.py", data_files=rel_datafiles)
58
59
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
736 schema_dict[field.name] = Value(str(field.type))
737
--> 738 parse_schema(writer.schema, features)
739 self.info.features = Features(features)
740
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in parse_schema(schema, schema_dict)
734 parse_schema(field.type.value_type, schema_dict[field.name])
735 else:
--> 736 schema_dict[field.name] = Value(str(field.type))
737
738 parse_schema(writer.schema, features)
<string> in __init__(self, dtype, id, _type)
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in __post_init__(self)
55
56 def __post_init__(self):
---> 57 self.pa_type = string_to_arrow(self.dtype)
58
59 def __call__(self):
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in string_to_arrow(type_str)
32 if str(type_str + "_") not in pa.__dict__:
33 raise ValueError(
---> 34 f"Neither {type_str} nor {type_str + '_'} seems to be a pyarrow data type. "
35 f"Please make sure to use a correct data type, see: "
36 f"https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions"
ValueError: Neither list<item: string> nor list<item: string>_ seems to be a pyarrow data type. Please make sure to use a correct data type, see: https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions
```
If I create the dataset imperatively, using a pyarrow table, the dataset is created correctly. If I override the `_prepare_split` method to avoid calling the validate schema, the dataset can load as well.
Hi, it depends on what it is in your `dataset_builder.py` file. Can you share it?
If you are just loading `json` files, you can also directly use the `json` script (which will find the schema/features from your JSON structure):
```python
from nlp import load_dataset
ds = load_dataset("json", data_files=rel_datafiles)
``` |
https://github.com/huggingface/datasets/issues/359 | ArrowBasedBuilder _prepare_split parse_schema breaks on nested structures | The behavior I'm seeing is from the `json` script.
I hacked this together to overcome the error with the `JSON` dataloader
```
class DatasetBuilder(hf_nlp.ArrowBasedBuilder):
BUILDER_CONFIG_CLASS = BuilderConfig
def _info(self):
return DatasetInfo()
def _split_generators(self, dl_manager):
""" We handle string, list and dicts in datafiles
"""
if isinstance(self.config.data_files, (str, list, tuple)):
files = self.config.data_files
if isinstance(files, str):
files = [files]
return [SplitGenerator(name=Split.TRAIN, gen_kwargs={"files": files})]
splits = []
for split_name in [Split.TRAIN, Split.VALIDATION, Split.TEST]:
if split_name in self.config.data_files:
files = self.config.data_files[split_name]
if isinstance(files, str):
files = [files]
splits.append(SplitGenerator(name=split_name, gen_kwargs={"files": files}))
return splits
def _prepare_split(self, split_generator):
fname = "{}-{}.arrow".format(self.name, split_generator.name)
fpath = os.path.join(self._cache_dir, fname)
writer = ArrowWriter(path=fpath)
generator = self._generate_tables(**split_generator.gen_kwargs)
for key, table in utils.tqdm(generator, unit=" tables", leave=False):
writer.write_table(table)
num_examples, num_bytes = writer.finalize()
split_generator.split_info.num_examples = num_examples
split_generator.split_info.num_bytes = num_bytes
# this is where the error is coming from
# def parse_schema(schema, schema_dict):
# for field in schema:
# if pa.types.is_struct(field.type):
# schema_dict[field.name] = {}
# parse_schema(field.type, schema_dict[field.name])
# elif pa.types.is_list(field.type) and pa.types.is_struct(field.type.value_type):
# schema_dict[field.name] = {}
# parse_schema(field.type.value_type, schema_dict[field.name])
# else:
# schema_dict[field.name] = Value(str(field.type))
#
# parse_schema(writer.schema, features)
# self.info.features = Features(features)
def _generate_tables(self, files):
for i, file in enumerate(files):
pa_table = paj.read_json(
file
)
yield i, pa_table
```
So I basically just don't populate the `self.info.features` though this doesn't seem to cause any problems in my downstream applications.
The other workaround I was doing was to just use pyarrow.json to build a table and then to create the Dataset with its constructor or from_table methods. `load_dataset` has nice split logic, so I'd prefer to use that.
| I tried using the Json dataloader to load some JSON lines files. but get an exception in the parse_schema function.
```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-23-9aecfbee53bd> in <module>
55 from nlp import load_dataset
56
---> 57 ds = load_dataset("../text2struct/model/dataset_builder.py", data_files=rel_datafiles)
58
59
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
736 schema_dict[field.name] = Value(str(field.type))
737
--> 738 parse_schema(writer.schema, features)
739 self.info.features = Features(features)
740
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in parse_schema(schema, schema_dict)
734 parse_schema(field.type.value_type, schema_dict[field.name])
735 else:
--> 736 schema_dict[field.name] = Value(str(field.type))
737
738 parse_schema(writer.schema, features)
<string> in __init__(self, dtype, id, _type)
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in __post_init__(self)
55
56 def __post_init__(self):
---> 57 self.pa_type = string_to_arrow(self.dtype)
58
59 def __call__(self):
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in string_to_arrow(type_str)
32 if str(type_str + "_") not in pa.__dict__:
33 raise ValueError(
---> 34 f"Neither {type_str} nor {type_str + '_'} seems to be a pyarrow data type. "
35 f"Please make sure to use a correct data type, see: "
36 f"https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions"
ValueError: Neither list<item: string> nor list<item: string>_ seems to be a pyarrow data type. Please make sure to use a correct data type, see: https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions
```
If I create the dataset imperatively, using a pyarrow table, the dataset is created correctly. If I override the `_prepare_split` method to avoid calling the validate schema, the dataset can load as well. | 254 | ArrowBasedBuilder _prepare_split parse_schema breaks on nested structures
I tried using the Json dataloader to load some JSON lines files. but get an exception in the parse_schema function.
```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-23-9aecfbee53bd> in <module>
55 from nlp import load_dataset
56
---> 57 ds = load_dataset("../text2struct/model/dataset_builder.py", data_files=rel_datafiles)
58
59
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
736 schema_dict[field.name] = Value(str(field.type))
737
--> 738 parse_schema(writer.schema, features)
739 self.info.features = Features(features)
740
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in parse_schema(schema, schema_dict)
734 parse_schema(field.type.value_type, schema_dict[field.name])
735 else:
--> 736 schema_dict[field.name] = Value(str(field.type))
737
738 parse_schema(writer.schema, features)
<string> in __init__(self, dtype, id, _type)
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in __post_init__(self)
55
56 def __post_init__(self):
---> 57 self.pa_type = string_to_arrow(self.dtype)
58
59 def __call__(self):
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in string_to_arrow(type_str)
32 if str(type_str + "_") not in pa.__dict__:
33 raise ValueError(
---> 34 f"Neither {type_str} nor {type_str + '_'} seems to be a pyarrow data type. "
35 f"Please make sure to use a correct data type, see: "
36 f"https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions"
ValueError: Neither list<item: string> nor list<item: string>_ seems to be a pyarrow data type. Please make sure to use a correct data type, see: https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions
```
If I create the dataset imperatively, using a pyarrow table, the dataset is created correctly. If I override the `_prepare_split` method to avoid calling the validate schema, the dataset can load as well.
The behavior I'm seeing is from the `json` script.
I hacked this together to overcome the error with the `JSON` dataloader
```
class DatasetBuilder(hf_nlp.ArrowBasedBuilder):
BUILDER_CONFIG_CLASS = BuilderConfig
def _info(self):
return DatasetInfo()
def _split_generators(self, dl_manager):
""" We handle string, list and dicts in datafiles
"""
if isinstance(self.config.data_files, (str, list, tuple)):
files = self.config.data_files
if isinstance(files, str):
files = [files]
return [SplitGenerator(name=Split.TRAIN, gen_kwargs={"files": files})]
splits = []
for split_name in [Split.TRAIN, Split.VALIDATION, Split.TEST]:
if split_name in self.config.data_files:
files = self.config.data_files[split_name]
if isinstance(files, str):
files = [files]
splits.append(SplitGenerator(name=split_name, gen_kwargs={"files": files}))
return splits
def _prepare_split(self, split_generator):
fname = "{}-{}.arrow".format(self.name, split_generator.name)
fpath = os.path.join(self._cache_dir, fname)
writer = ArrowWriter(path=fpath)
generator = self._generate_tables(**split_generator.gen_kwargs)
for key, table in utils.tqdm(generator, unit=" tables", leave=False):
writer.write_table(table)
num_examples, num_bytes = writer.finalize()
split_generator.split_info.num_examples = num_examples
split_generator.split_info.num_bytes = num_bytes
# this is where the error is coming from
# def parse_schema(schema, schema_dict):
# for field in schema:
# if pa.types.is_struct(field.type):
# schema_dict[field.name] = {}
# parse_schema(field.type, schema_dict[field.name])
# elif pa.types.is_list(field.type) and pa.types.is_struct(field.type.value_type):
# schema_dict[field.name] = {}
# parse_schema(field.type.value_type, schema_dict[field.name])
# else:
# schema_dict[field.name] = Value(str(field.type))
#
# parse_schema(writer.schema, features)
# self.info.features = Features(features)
def _generate_tables(self, files):
for i, file in enumerate(files):
pa_table = paj.read_json(
file
)
yield i, pa_table
```
So I basically just don't populate the `self.info.features` though this doesn't seem to cause any problems in my downstream applications.
The other workaround I was doing was to just use pyarrow.json to build a table and then to create the Dataset with its constructor or from_table methods. `load_dataset` has nice split logic, so I'd prefer to use that.
|
https://github.com/huggingface/datasets/issues/359 | ArrowBasedBuilder _prepare_split parse_schema breaks on nested structures | Also noticed that if you for example in a loader script
```
from nlp import ArrowBasedBuilder
class MyBuilder(ArrowBasedBuilder):
...
```
and use that in the subclass, it will be on the module's __dict__ and will be selected before the `MyBuilder` subclass, and it will raise `NotImplementedError` on its `_generate_examples` method... In the code it check for abstract classes but Builder and ArrowBasedBuilder aren't abstract classes, they're regular classes with `@abstract_methods`. | I tried using the Json dataloader to load some JSON lines files. but get an exception in the parse_schema function.
```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-23-9aecfbee53bd> in <module>
55 from nlp import load_dataset
56
---> 57 ds = load_dataset("../text2struct/model/dataset_builder.py", data_files=rel_datafiles)
58
59
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
736 schema_dict[field.name] = Value(str(field.type))
737
--> 738 parse_schema(writer.schema, features)
739 self.info.features = Features(features)
740
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in parse_schema(schema, schema_dict)
734 parse_schema(field.type.value_type, schema_dict[field.name])
735 else:
--> 736 schema_dict[field.name] = Value(str(field.type))
737
738 parse_schema(writer.schema, features)
<string> in __init__(self, dtype, id, _type)
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in __post_init__(self)
55
56 def __post_init__(self):
---> 57 self.pa_type = string_to_arrow(self.dtype)
58
59 def __call__(self):
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in string_to_arrow(type_str)
32 if str(type_str + "_") not in pa.__dict__:
33 raise ValueError(
---> 34 f"Neither {type_str} nor {type_str + '_'} seems to be a pyarrow data type. "
35 f"Please make sure to use a correct data type, see: "
36 f"https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions"
ValueError: Neither list<item: string> nor list<item: string>_ seems to be a pyarrow data type. Please make sure to use a correct data type, see: https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions
```
If I create the dataset imperatively, using a pyarrow table, the dataset is created correctly. If I override the `_prepare_split` method to avoid calling the validate schema, the dataset can load as well. | 70 | ArrowBasedBuilder _prepare_split parse_schema breaks on nested structures
I tried using the Json dataloader to load some JSON lines files. but get an exception in the parse_schema function.
```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-23-9aecfbee53bd> in <module>
55 from nlp import load_dataset
56
---> 57 ds = load_dataset("../text2struct/model/dataset_builder.py", data_files=rel_datafiles)
58
59
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
736 schema_dict[field.name] = Value(str(field.type))
737
--> 738 parse_schema(writer.schema, features)
739 self.info.features = Features(features)
740
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in parse_schema(schema, schema_dict)
734 parse_schema(field.type.value_type, schema_dict[field.name])
735 else:
--> 736 schema_dict[field.name] = Value(str(field.type))
737
738 parse_schema(writer.schema, features)
<string> in __init__(self, dtype, id, _type)
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in __post_init__(self)
55
56 def __post_init__(self):
---> 57 self.pa_type = string_to_arrow(self.dtype)
58
59 def __call__(self):
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in string_to_arrow(type_str)
32 if str(type_str + "_") not in pa.__dict__:
33 raise ValueError(
---> 34 f"Neither {type_str} nor {type_str + '_'} seems to be a pyarrow data type. "
35 f"Please make sure to use a correct data type, see: "
36 f"https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions"
ValueError: Neither list<item: string> nor list<item: string>_ seems to be a pyarrow data type. Please make sure to use a correct data type, see: https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions
```
If I create the dataset imperatively, using a pyarrow table, the dataset is created correctly. If I override the `_prepare_split` method to avoid calling the validate schema, the dataset can load as well.
Also noticed that if you for example in a loader script
```
from nlp import ArrowBasedBuilder
class MyBuilder(ArrowBasedBuilder):
...
```
and use that in the subclass, it will be on the module's __dict__ and will be selected before the `MyBuilder` subclass, and it will raise `NotImplementedError` on its `_generate_examples` method... In the code it check for abstract classes but Builder and ArrowBasedBuilder aren't abstract classes, they're regular classes with `@abstract_methods`. |
https://github.com/huggingface/datasets/issues/359 | ArrowBasedBuilder _prepare_split parse_schema breaks on nested structures | Indeed this is part of a more general limitation which is the fact that we should generate and update the `features` from the auto-inferred Arrow schema when they are not provided (also happen when a user change the schema using `map()`, the features should be auto-generated and guessed as much as possible to keep the `features` synced with the underlying Arrow table schema).
We will try to solve this soon. | I tried using the Json dataloader to load some JSON lines files. but get an exception in the parse_schema function.
```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-23-9aecfbee53bd> in <module>
55 from nlp import load_dataset
56
---> 57 ds = load_dataset("../text2struct/model/dataset_builder.py", data_files=rel_datafiles)
58
59
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
736 schema_dict[field.name] = Value(str(field.type))
737
--> 738 parse_schema(writer.schema, features)
739 self.info.features = Features(features)
740
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in parse_schema(schema, schema_dict)
734 parse_schema(field.type.value_type, schema_dict[field.name])
735 else:
--> 736 schema_dict[field.name] = Value(str(field.type))
737
738 parse_schema(writer.schema, features)
<string> in __init__(self, dtype, id, _type)
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in __post_init__(self)
55
56 def __post_init__(self):
---> 57 self.pa_type = string_to_arrow(self.dtype)
58
59 def __call__(self):
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in string_to_arrow(type_str)
32 if str(type_str + "_") not in pa.__dict__:
33 raise ValueError(
---> 34 f"Neither {type_str} nor {type_str + '_'} seems to be a pyarrow data type. "
35 f"Please make sure to use a correct data type, see: "
36 f"https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions"
ValueError: Neither list<item: string> nor list<item: string>_ seems to be a pyarrow data type. Please make sure to use a correct data type, see: https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions
```
If I create the dataset imperatively, using a pyarrow table, the dataset is created correctly. If I override the `_prepare_split` method to avoid calling the validate schema, the dataset can load as well. | 70 | ArrowBasedBuilder _prepare_split parse_schema breaks on nested structures
I tried using the Json dataloader to load some JSON lines files. but get an exception in the parse_schema function.
```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-23-9aecfbee53bd> in <module>
55 from nlp import load_dataset
56
---> 57 ds = load_dataset("../text2struct/model/dataset_builder.py", data_files=rel_datafiles)
58
59
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
736 schema_dict[field.name] = Value(str(field.type))
737
--> 738 parse_schema(writer.schema, features)
739 self.info.features = Features(features)
740
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in parse_schema(schema, schema_dict)
734 parse_schema(field.type.value_type, schema_dict[field.name])
735 else:
--> 736 schema_dict[field.name] = Value(str(field.type))
737
738 parse_schema(writer.schema, features)
<string> in __init__(self, dtype, id, _type)
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in __post_init__(self)
55
56 def __post_init__(self):
---> 57 self.pa_type = string_to_arrow(self.dtype)
58
59 def __call__(self):
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in string_to_arrow(type_str)
32 if str(type_str + "_") not in pa.__dict__:
33 raise ValueError(
---> 34 f"Neither {type_str} nor {type_str + '_'} seems to be a pyarrow data type. "
35 f"Please make sure to use a correct data type, see: "
36 f"https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions"
ValueError: Neither list<item: string> nor list<item: string>_ seems to be a pyarrow data type. Please make sure to use a correct data type, see: https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions
```
If I create the dataset imperatively, using a pyarrow table, the dataset is created correctly. If I override the `_prepare_split` method to avoid calling the validate schema, the dataset can load as well.
Indeed this is part of a more general limitation which is the fact that we should generate and update the `features` from the auto-inferred Arrow schema when they are not provided (also happen when a user change the schema using `map()`, the features should be auto-generated and guessed as much as possible to keep the `features` synced with the underlying Arrow table schema).
We will try to solve this soon. |
https://github.com/huggingface/datasets/issues/355 | can't load SNLI dataset | I just added the processed files of `snli` on our google storage, so that when you do `load_dataset` it can download the processed files from there :)
We are thinking about having available those processed files for more datasets in the future, because sometimes files aren't available (like for `snli`), or the download speed is too slow, or sometimes the files take time to be processed. | `nlp` seems to load `snli` from some URL based on nlp.stanford.edu. This subdomain is frequently down -- including right now, when I'd like to load `snli` in a Colab notebook, but can't.
Is there a plan to move these datasets to huggingface servers for a more stable solution?
Btw, here's the stack trace:
```
File "/content/nlp/src/nlp/builder.py", line 432, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/content/nlp/src/nlp/builder.py", line 466, in _download_and_prepare
split_generators = self._split_generators(dl_manager, **split_generators_kwargs)
File "/content/nlp/src/nlp/datasets/snli/e417f6f2e16254938d977a17ed32f3998f5b23e4fcab0f6eb1d28784f23ea60d/snli.py", line 76, in _split_generators
dl_dir = dl_manager.download_and_extract(_DATA_URL)
File "/content/nlp/src/nlp/utils/download_manager.py", line 217, in download_and_extract
return self.extract(self.download(url_or_urls))
File "/content/nlp/src/nlp/utils/download_manager.py", line 156, in download
lambda url: cached_path(url, download_config=self._download_config,), url_or_urls,
File "/content/nlp/src/nlp/utils/py_utils.py", line 190, in map_nested
return function(data_struct)
File "/content/nlp/src/nlp/utils/download_manager.py", line 156, in <lambda>
lambda url: cached_path(url, download_config=self._download_config,), url_or_urls,
File "/content/nlp/src/nlp/utils/file_utils.py", line 198, in cached_path
local_files_only=download_config.local_files_only,
File "/content/nlp/src/nlp/utils/file_utils.py", line 356, in get_from_cache
raise ConnectionError("Couldn't reach {}".format(url))
ConnectionError: Couldn't reach https://nlp.stanford.edu/projects/snli/snli_1.0.zip
``` | 66 | can't load SNLI dataset
`nlp` seems to load `snli` from some URL based on nlp.stanford.edu. This subdomain is frequently down -- including right now, when I'd like to load `snli` in a Colab notebook, but can't.
Is there a plan to move these datasets to huggingface servers for a more stable solution?
Btw, here's the stack trace:
```
File "/content/nlp/src/nlp/builder.py", line 432, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/content/nlp/src/nlp/builder.py", line 466, in _download_and_prepare
split_generators = self._split_generators(dl_manager, **split_generators_kwargs)
File "/content/nlp/src/nlp/datasets/snli/e417f6f2e16254938d977a17ed32f3998f5b23e4fcab0f6eb1d28784f23ea60d/snli.py", line 76, in _split_generators
dl_dir = dl_manager.download_and_extract(_DATA_URL)
File "/content/nlp/src/nlp/utils/download_manager.py", line 217, in download_and_extract
return self.extract(self.download(url_or_urls))
File "/content/nlp/src/nlp/utils/download_manager.py", line 156, in download
lambda url: cached_path(url, download_config=self._download_config,), url_or_urls,
File "/content/nlp/src/nlp/utils/py_utils.py", line 190, in map_nested
return function(data_struct)
File "/content/nlp/src/nlp/utils/download_manager.py", line 156, in <lambda>
lambda url: cached_path(url, download_config=self._download_config,), url_or_urls,
File "/content/nlp/src/nlp/utils/file_utils.py", line 198, in cached_path
local_files_only=download_config.local_files_only,
File "/content/nlp/src/nlp/utils/file_utils.py", line 356, in get_from_cache
raise ConnectionError("Couldn't reach {}".format(url))
ConnectionError: Couldn't reach https://nlp.stanford.edu/projects/snli/snli_1.0.zip
```
I just added the processed files of `snli` on our google storage, so that when you do `load_dataset` it can download the processed files from there :)
We are thinking about having available those processed files for more datasets in the future, because sometimes files aren't available (like for `snli`), or the download speed is too slow, or sometimes the files take time to be processed. |
https://github.com/huggingface/datasets/issues/353 | [Dataset requests] New datasets for Text Classification | - `nlp` has MR! It's called `rotten_tomatoes`
- SST is part of GLUE, or is that just SST-2?
- `nlp` also has `ag_news`, a popular news classification dataset
I'd also like to see:
- the Yahoo Answers topic classification dataset
- the Kaggle Fake News classification dataset | We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705 | 47 | [Dataset requests] New datasets for Text Classification
We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705
- `nlp` has MR! It's called `rotten_tomatoes`
- SST is part of GLUE, or is that just SST-2?
- `nlp` also has `ag_news`, a popular news classification dataset
I'd also like to see:
- the Yahoo Answers topic classification dataset
- the Kaggle Fake News classification dataset |
https://github.com/huggingface/datasets/issues/353 | [Dataset requests] New datasets for Text Classification | Thanks @jxmorris12 for pointing this out.
In glue we only have SST-2 maybe we can add separately SST-1.
| We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705 | 18 | [Dataset requests] New datasets for Text Classification
We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705
Thanks @jxmorris12 for pointing this out.
In glue we only have SST-2 maybe we can add separately SST-1.
|
https://github.com/huggingface/datasets/issues/353 | [Dataset requests] New datasets for Text Classification | This is the homepage for the Amazon dataset: https://www.kaggle.com/datafiniti/consumer-reviews-of-amazon-products
Is there an easy way to download kaggle datasets programmatically? If so, I can add this one! | We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705 | 26 | [Dataset requests] New datasets for Text Classification
We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705
This is the homepage for the Amazon dataset: https://www.kaggle.com/datafiniti/consumer-reviews-of-amazon-products
Is there an easy way to download kaggle datasets programmatically? If so, I can add this one! |
https://github.com/huggingface/datasets/issues/353 | [Dataset requests] New datasets for Text Classification | Hi @jxmorris12 for now I think our `dl_manager` does not download from Kaggle.
@thomwolf , @lhoestq | We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705 | 16 | [Dataset requests] New datasets for Text Classification
We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705
Hi @jxmorris12 for now I think our `dl_manager` does not download from Kaggle.
@thomwolf , @lhoestq |
https://github.com/huggingface/datasets/issues/353 | [Dataset requests] New datasets for Text Classification | Great list. Any idea if Amazon Reviews has been added?
- ~40 GB of text (sadly no emoji)
- popular MLM pre-training dataset before bigger datasets like WebText https://arxiv.org/abs/1808.01371
- turns out that binarizing the 1-5 star rating leads to great Pos/Neg/Neutral dataset, T5 paper claims to get very high accuracy (98%!) on this with small amount of finetuning https://arxiv.org/abs/2004.14546
Apologies if it's been included (great to see where) and if not, it's one of the better medium/large NLP dataset for semi-supervised learning, albeit a bit out of date.
Thanks!!
cc @sshleifer | We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705 | 92 | [Dataset requests] New datasets for Text Classification
We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705
Great list. Any idea if Amazon Reviews has been added?
- ~40 GB of text (sadly no emoji)
- popular MLM pre-training dataset before bigger datasets like WebText https://arxiv.org/abs/1808.01371
- turns out that binarizing the 1-5 star rating leads to great Pos/Neg/Neutral dataset, T5 paper claims to get very high accuracy (98%!) on this with small amount of finetuning https://arxiv.org/abs/2004.14546
Apologies if it's been included (great to see where) and if not, it's one of the better medium/large NLP dataset for semi-supervised learning, albeit a bit out of date.
Thanks!!
cc @sshleifer |
https://github.com/huggingface/datasets/issues/353 | [Dataset requests] New datasets for Text Classification | On the Amazon Reviews dataset, the original UCSD website has noted these are now updated to include product reviews through 2018 -- actually quite recent compared to many other datasets. Almost certainly the largest NLP dataset out there with labels!
https://jmcauley.ucsd.edu/data/amazon/
Any chance someone has time to onboard this dataset in a HF way?
cc @sshleifer | We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705 | 56 | [Dataset requests] New datasets for Text Classification
We are missing a few datasets for Text Classification which is an important field.
Namely, it would be really nice to add:
- [x] TREC-6 dataset (see here for instance: https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html#torchnlp.datasets.trec_dataset) **[done]**
- #386
- [x] Yelp-5
- #1315
- [x] Movie review (Movie Review (MR) dataset [156]) **[done (same as rotten_tomatoes)]**
- [x] SST (Stanford Sentiment Treebank) **[include in glue]**
- #1934
- [ ] Multi-Perspective Question Answering (MPQA) dataset **[require authentication (indeed manual download)]**
- [x] Amazon. This is a popular corpus of product reviews collected from the Amazon website [159]. It contains labels for both binary classification and multi-class (5-class) classification
- #791
- #1389
- [x] 20 Newsgroups. The 20 Newsgroups dataset **[done]**
- #410
- [x] Sogou News dataset **[done]**
- #450
- [x] Reuters news. The Reuters-21578 dataset [165] **[done]**
- #471
- [x] DBpedia. The DBpedia dataset [170]
- #1116
- [ ] Ohsumed. The Ohsumed collection [171] is a subset of the MEDLINE database
- [ ] EUR-Lex. The EUR-Lex dataset
- [x] WOS. The Web Of Science (WOS) dataset **[done]**
- #424
- [ ] PubMed. PubMed [173]
- [x] TREC-QA: TREC-6 + TREC-50
- See above: TREC-6 dataset
- [x] Quora. The Quora dataset [180]
- #366
All these datasets are cited in https://arxiv.org/abs/2004.03705
On the Amazon Reviews dataset, the original UCSD website has noted these are now updated to include product reviews through 2018 -- actually quite recent compared to many other datasets. Almost certainly the largest NLP dataset out there with labels!
https://jmcauley.ucsd.edu/data/amazon/
Any chance someone has time to onboard this dataset in a HF way?
cc @sshleifer |
https://github.com/huggingface/datasets/issues/347 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa') | This is probably a Windows issue, we need to specify the encoding when `load_dataset()` reads the original CSV file.
Try to find the `open()` statement called by `load_dataset()` and add an `encoding='utf-8'` parameter.
See issues #242 and #307 | 
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
| 38 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa')
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
This is probably a Windows issue, we need to specify the encoding when `load_dataset()` reads the original CSV file.
Try to find the `open()` statement called by `load_dataset()` and add an `encoding='utf-8'` parameter.
See issues #242 and #307 |
https://github.com/huggingface/datasets/issues/347 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa') | It should be in `xtreme.py:L755`:
```python
if self.config.name == "tydiqa" or self.config.name.startswith("MLQA") or self.config.name == "SQuAD":
with open(filepath) as f:
data = json.load(f)
```
Could you try to add the encoding parameter:
```python
open(filepath, encoding='utf-8')
``` | 
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
| 36 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa')
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
It should be in `xtreme.py:L755`:
```python
if self.config.name == "tydiqa" or self.config.name.startswith("MLQA") or self.config.name == "SQuAD":
with open(filepath) as f:
data = json.load(f)
```
Could you try to add the encoding parameter:
```python
open(filepath, encoding='utf-8')
``` |
https://github.com/huggingface/datasets/issues/347 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa') | Hello @jerryIsHere :) Did it work ?
If so we may change the dataset script to force the utf-8 encoding | 
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
| 20 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa')
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
Hello @jerryIsHere :) Did it work ?
If so we may change the dataset script to force the utf-8 encoding |
https://github.com/huggingface/datasets/issues/347 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa') | @lhoestq sorry for being that late, I found 4 copy of xtreme.py. I did the changes as what has been told to all of them.
The problem is not solved | 
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
| 30 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa')
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
@lhoestq sorry for being that late, I found 4 copy of xtreme.py. I did the changes as what has been told to all of them.
The problem is not solved |
https://github.com/huggingface/datasets/issues/347 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa') | Could you provide a better error message so that we can make sure it comes from the opening of the `tydiqa`'s json files ?
| 
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
| 24 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa')
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
Could you provide a better error message so that we can make sure it comes from the opening of the `tydiqa`'s json files ?
|
https://github.com/huggingface/datasets/issues/347 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa') | @lhoestq
The error message is same as before:
Exception has occurred: UnicodeDecodeError
'cp950' codec can't decode byte 0xe2 in position 111: illegal multibyte sequence
File "D:\python\test\test.py", line 3, in <module>
dataset = load_dataset('xtreme', 'tydiqa')
I said that I found 4 copy of xtreme.py and add the 「, encoding='utf-8'」 parameter to the open() function
these python script was found under this directory
C:\Users\USER\AppData\Local\Programs\Python\Python37\Lib\site-packages\nlp\datasets\xtreme
| 
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
| 63 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa')
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
@lhoestq
The error message is same as before:
Exception has occurred: UnicodeDecodeError
'cp950' codec can't decode byte 0xe2 in position 111: illegal multibyte sequence
File "D:\python\test\test.py", line 3, in <module>
dataset = load_dataset('xtreme', 'tydiqa')
I said that I found 4 copy of xtreme.py and add the 「, encoding='utf-8'」 parameter to the open() function
these python script was found under this directory
C:\Users\USER\AppData\Local\Programs\Python\Python37\Lib\site-packages\nlp\datasets\xtreme
|
https://github.com/huggingface/datasets/issues/347 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa') | Hi there !
I encountered the same issue with the IMDB dataset on windows. It threw an error about charmap not being able to decode a symbol during the first time I tried to download it. I checked on a remote linux machine I have, and it can't be reproduced.
I added ```encoding='UTF-8'``` to both lines that have ```open``` in ```imdb.py``` (108 and 114) and it worked for me.
Thank you ! | 
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
| 72 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa')
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
Hi there !
I encountered the same issue with the IMDB dataset on windows. It threw an error about charmap not being able to decode a symbol during the first time I tried to download it. I checked on a remote linux machine I have, and it can't be reproduced.
I added ```encoding='UTF-8'``` to both lines that have ```open``` in ```imdb.py``` (108 and 114) and it worked for me.
Thank you ! |
https://github.com/huggingface/datasets/issues/347 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa') | > Hi there !
> I encountered the same issue with the IMDB dataset on windows. It threw an error about charmap not being able to decode a symbol during the first time I tried to download it. I checked on a remote linux machine I have, and it can't be reproduced.
> I added `encoding='UTF-8'` to both lines that have `open` in `imdb.py` (108 and 114) and it worked for me.
> Thank you !
Hello !
Glad you managed to fix this issue on your side.
Do you mind opening a PR for IMDB ? | 
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
| 97 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa')
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
> Hi there !
> I encountered the same issue with the IMDB dataset on windows. It threw an error about charmap not being able to decode a symbol during the first time I tried to download it. I checked on a remote linux machine I have, and it can't be reproduced.
> I added `encoding='UTF-8'` to both lines that have `open` in `imdb.py` (108 and 114) and it worked for me.
> Thank you !
Hello !
Glad you managed to fix this issue on your side.
Do you mind opening a PR for IMDB ? |
https://github.com/huggingface/datasets/issues/347 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa') | > This is probably a Windows issue, we need to specify the encoding when `load_dataset()` reads the original CSV file.
> Try to find the `open()` statement called by `load_dataset()` and add an `encoding='utf-8'` parameter.
> See issues #242 and #307
Sorry for not responding for about a month.
I have just found that it is necessary to change / add the environment variable as what was told in #242.
Everything works after I add the new environment variable and restart my PC.
I think the encoding issue for windows isn't limited to the open() function call specific to few dataset, but actually in the entire library, depends on the machine / os you use. | 
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
| 115 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa')
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
> This is probably a Windows issue, we need to specify the encoding when `load_dataset()` reads the original CSV file.
> Try to find the `open()` statement called by `load_dataset()` and add an `encoding='utf-8'` parameter.
> See issues #242 and #307
Sorry for not responding for about a month.
I have just found that it is necessary to change / add the environment variable as what was told in #242.
Everything works after I add the new environment variable and restart my PC.
I think the encoding issue for windows isn't limited to the open() function call specific to few dataset, but actually in the entire library, depends on the machine / os you use. |
https://github.com/huggingface/datasets/issues/347 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa') | Since #481 we shouldn't have other issues with encodings as they need to be set to "utf-8" be default.
Closing this one, but feel free to re-open if you gave other questions | 
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
| 32 | 'cp950' codec error from load_dataset('xtreme', 'tydiqa')
I guess the error is related to python source encoding issue that my PC is trying to decode the source code with wrong encoding-decoding tools, perhaps :
https://www.python.org/dev/peps/pep-0263/
I guess the error was triggered by the code " module = importlib.import_module(module_path)" at line 57 in the source code: nlp/src/nlp/load.py / (https://github.com/huggingface/nlp/blob/911d5596f9b500e39af8642fe3d1b891758999c7/src/nlp/load.py#L51)
Any ideas?
p.s. tried the same code on colab, that runs perfectly
Since #481 we shouldn't have other issues with encodings as they need to be set to "utf-8" be default.
Closing this one, but feel free to re-open if you gave other questions |
https://github.com/huggingface/datasets/issues/345 | Supporting documents in ELI5 | Hi @saverymax ! For licensing reasons, the original team was unable to release pre-processed CommonCrawl documents. Instead, they provided a script to re-create them from a CommonCrawl dump, but it unfortunately requires access to a medium-large size cluster:
https://github.com/facebookresearch/ELI5#downloading-support-documents-from-the-commoncrawl
In order to make the task accessible to people who may not have access to this kind of infrastructure, we suggest to use Wikipedia as a knowledge source rather than the full CommonCrawl. The following blog post shows how you can create Wikipedia support documents and get a performance that is on par with a system that uses CommonCrawl pages.
https://yjernite.github.io/lfqa.html#task_description
Hope that helps, using ElasticSearch to index Wiki40b and create the documents should take about 4 hours. Let us know if you have any trouble with the blog post though! | I was attempting to use the ELI5 dataset, when I realized that huggingface does not provide the supporting documents (the source documents from the common crawl). Without the supporting documents, this makes the dataset about as useful for my project as a block of cheese, or some other more apt metaphor. According to facebook, the entire document collection is quite large. However, it would still be helpful to at least include a subset of the supporting documents i.e., having some data is better than having a block of cheese, in my case at least.
If you choose not to include them, it would be helpful to have documentation mentioning this specifically. It is especially confusing because the hf nlp ELI5 dataset has the key `'document'` but there are no documents to be found :( | 130 | Supporting documents in ELI5
I was attempting to use the ELI5 dataset, when I realized that huggingface does not provide the supporting documents (the source documents from the common crawl). Without the supporting documents, this makes the dataset about as useful for my project as a block of cheese, or some other more apt metaphor. According to facebook, the entire document collection is quite large. However, it would still be helpful to at least include a subset of the supporting documents i.e., having some data is better than having a block of cheese, in my case at least.
If you choose not to include them, it would be helpful to have documentation mentioning this specifically. It is especially confusing because the hf nlp ELI5 dataset has the key `'document'` but there are no documents to be found :(
Hi @saverymax ! For licensing reasons, the original team was unable to release pre-processed CommonCrawl documents. Instead, they provided a script to re-create them from a CommonCrawl dump, but it unfortunately requires access to a medium-large size cluster:
https://github.com/facebookresearch/ELI5#downloading-support-documents-from-the-commoncrawl
In order to make the task accessible to people who may not have access to this kind of infrastructure, we suggest to use Wikipedia as a knowledge source rather than the full CommonCrawl. The following blog post shows how you can create Wikipedia support documents and get a performance that is on par with a system that uses CommonCrawl pages.
https://yjernite.github.io/lfqa.html#task_description
Hope that helps, using ElasticSearch to index Wiki40b and create the documents should take about 4 hours. Let us know if you have any trouble with the blog post though! |
https://github.com/huggingface/datasets/issues/345 | Supporting documents in ELI5 | Hi, thanks for the quick response. The blog post is quite an interesting working example, thanks for sharing it.
Two follow-up points/questions about my original question:
1. Yes, I read that the facebook team could not share the CommonCrawl b/c of licensing reasons. They state "No, we are not allowed to host processed Reddit or CommonCrawl data," which indicates they could also not share the Reddit data for licensing reasons. But it seems that HuggingFace is able to share the Reddit data, so why not a subset of CommonCrawl?
2. Thanks for the suggestion about ElasticSearch and Wiki40b. This is good to know about performance. I definitely could do the indexing and querying myself. What I like about the ELI5 dataset though, at least what is suggested by the paper, is that to create the dataset they had already selected the top 100 web sources and made a single support document from those. Though it doesn't appear to be too sophisticated an approach, having a single support document pre-computed (without having to run the facebook code or a replacement with another dataset) is super useful for my work, especially since I'm not working on developing the latest and greatest retrieval model. Of course, I don't expect HF NLP datasets to be perfectly tailored to my use-case. I know there is overhead to any project, I'm just illustrating a use-case of ELI5 which is not possible with the data provided as-is. If it's for licensing reasons, that is perfectly acceptable a reason, and I appreciate your response. | I was attempting to use the ELI5 dataset, when I realized that huggingface does not provide the supporting documents (the source documents from the common crawl). Without the supporting documents, this makes the dataset about as useful for my project as a block of cheese, or some other more apt metaphor. According to facebook, the entire document collection is quite large. However, it would still be helpful to at least include a subset of the supporting documents i.e., having some data is better than having a block of cheese, in my case at least.
If you choose not to include them, it would be helpful to have documentation mentioning this specifically. It is especially confusing because the hf nlp ELI5 dataset has the key `'document'` but there are no documents to be found :( | 256 | Supporting documents in ELI5
I was attempting to use the ELI5 dataset, when I realized that huggingface does not provide the supporting documents (the source documents from the common crawl). Without the supporting documents, this makes the dataset about as useful for my project as a block of cheese, or some other more apt metaphor. According to facebook, the entire document collection is quite large. However, it would still be helpful to at least include a subset of the supporting documents i.e., having some data is better than having a block of cheese, in my case at least.
If you choose not to include them, it would be helpful to have documentation mentioning this specifically. It is especially confusing because the hf nlp ELI5 dataset has the key `'document'` but there are no documents to be found :(
Hi, thanks for the quick response. The blog post is quite an interesting working example, thanks for sharing it.
Two follow-up points/questions about my original question:
1. Yes, I read that the facebook team could not share the CommonCrawl b/c of licensing reasons. They state "No, we are not allowed to host processed Reddit or CommonCrawl data," which indicates they could also not share the Reddit data for licensing reasons. But it seems that HuggingFace is able to share the Reddit data, so why not a subset of CommonCrawl?
2. Thanks for the suggestion about ElasticSearch and Wiki40b. This is good to know about performance. I definitely could do the indexing and querying myself. What I like about the ELI5 dataset though, at least what is suggested by the paper, is that to create the dataset they had already selected the top 100 web sources and made a single support document from those. Though it doesn't appear to be too sophisticated an approach, having a single support document pre-computed (without having to run the facebook code or a replacement with another dataset) is super useful for my work, especially since I'm not working on developing the latest and greatest retrieval model. Of course, I don't expect HF NLP datasets to be perfectly tailored to my use-case. I know there is overhead to any project, I'm just illustrating a use-case of ELI5 which is not possible with the data provided as-is. If it's for licensing reasons, that is perfectly acceptable a reason, and I appreciate your response. |
https://github.com/huggingface/datasets/issues/331 | Loading CNN/Daily Mail dataset produces `nlp.utils.info_utils.NonMatchingSplitsSizesError` | I couldn't reproduce on my side.
It looks like you were not able to generate all the examples, and you have the problem for each split train-test-validation.
Could you try to enable logging, try again and send the logs ?
```python
import logging
logging.basicConfig(level=logging.INFO)
``` | ```
>>> import nlp
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
``` | 45 | Loading CNN/Daily Mail dataset produces `nlp.utils.info_utils.NonMatchingSplitsSizesError`
```
>>> import nlp
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
```
I couldn't reproduce on my side.
It looks like you were not able to generate all the examples, and you have the problem for each split train-test-validation.
Could you try to enable logging, try again and send the logs ?
```python
import logging
logging.basicConfig(level=logging.INFO)
``` |
https://github.com/huggingface/datasets/issues/331 | Loading CNN/Daily Mail dataset produces `nlp.utils.info_utils.NonMatchingSplitsSizesError` | here's the log
```
>>> import nlp
import logging
logging.basicConfig(level=logging.INFO)
nlp.load_dataset('cnn_dailymail', '3.0.0')
>>> import logging
>>> logging.basicConfig(level=logging.INFO)
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
INFO:nlp.load:Checking /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py for additional imports.
INFO:filelock:Lock 140443095301136 acquired on /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py.lock
INFO:nlp.load:Found main folder for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail
INFO:nlp.load:Found specific version folder for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad
INFO:nlp.load:Found script file from https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py to /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/cnn_dailymail.py
INFO:nlp.load:Updating dataset infos file from https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/dataset_infos.json to /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/dataset_infos.json
INFO:nlp.load:Found metadata file for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/cnn_dailymail.json
INFO:filelock:Lock 140443095301136 released on /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py.lock
INFO:nlp.info:Loading Dataset Infos from /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad
INFO:nlp.builder:Generating dataset cnn_dailymail (/u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0)
INFO:nlp.builder:Dataset not on Hf google storage. Downloading and preparing it from source
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
INFO:nlp.utils.info_utils:All the checksums matched successfully.
INFO:nlp.builder:Generating split train
INFO:nlp.arrow_writer:Done writing 285161 examples in 1240618482 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-train.arrow.
INFO:nlp.builder:Generating split validation
INFO:nlp.arrow_writer:Done writing 13255 examples in 56637485 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-validation.arrow.
INFO:nlp.builder:Generating split test
INFO:nlp.arrow_writer:Done writing 11379 examples in 48931393 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-test.arrow.
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
``` | ```
>>> import nlp
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
``` | 223 | Loading CNN/Daily Mail dataset produces `nlp.utils.info_utils.NonMatchingSplitsSizesError`
```
>>> import nlp
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
```
here's the log
```
>>> import nlp
import logging
logging.basicConfig(level=logging.INFO)
nlp.load_dataset('cnn_dailymail', '3.0.0')
>>> import logging
>>> logging.basicConfig(level=logging.INFO)
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
INFO:nlp.load:Checking /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py for additional imports.
INFO:filelock:Lock 140443095301136 acquired on /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py.lock
INFO:nlp.load:Found main folder for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail
INFO:nlp.load:Found specific version folder for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad
INFO:nlp.load:Found script file from https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py to /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/cnn_dailymail.py
INFO:nlp.load:Updating dataset infos file from https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/dataset_infos.json to /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/dataset_infos.json
INFO:nlp.load:Found metadata file for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/cnn_dailymail.json
INFO:filelock:Lock 140443095301136 released on /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py.lock
INFO:nlp.info:Loading Dataset Infos from /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad
INFO:nlp.builder:Generating dataset cnn_dailymail (/u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0)
INFO:nlp.builder:Dataset not on Hf google storage. Downloading and preparing it from source
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
INFO:nlp.utils.info_utils:All the checksums matched successfully.
INFO:nlp.builder:Generating split train
INFO:nlp.arrow_writer:Done writing 285161 examples in 1240618482 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-train.arrow.
INFO:nlp.builder:Generating split validation
INFO:nlp.arrow_writer:Done writing 13255 examples in 56637485 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-validation.arrow.
INFO:nlp.builder:Generating split test
INFO:nlp.arrow_writer:Done writing 11379 examples in 48931393 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-test.arrow.
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
``` |
https://github.com/huggingface/datasets/issues/331 | Loading CNN/Daily Mail dataset produces `nlp.utils.info_utils.NonMatchingSplitsSizesError` | > here's the log
>
> ```
> >>> import nlp
> import logging
> logging.basicConfig(level=logging.INFO)
> nlp.load_dataset('cnn_dailymail', '3.0.0')
> >>> import logging
> >>> logging.basicConfig(level=logging.INFO)
> >>> nlp.load_dataset('cnn_dailymail', '3.0.0')
> INFO:nlp.load:Checking /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py for additional imports.
> INFO:filelock:Lock 140443095301136 acquired on /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py.lock
> INFO:nlp.load:Found main folder for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail
> INFO:nlp.load:Found specific version folder for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad
> INFO:nlp.load:Found script file from https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py to /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/cnn_dailymail.py
> INFO:nlp.load:Updating dataset infos file from https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/dataset_infos.json to /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/dataset_infos.json
> INFO:nlp.load:Found metadata file for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/cnn_dailymail.json
> INFO:filelock:Lock 140443095301136 released on /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py.lock
> INFO:nlp.info:Loading Dataset Infos from /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad
> INFO:nlp.builder:Generating dataset cnn_dailymail (/u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0)
> INFO:nlp.builder:Dataset not on Hf google storage. Downloading and preparing it from source
> Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
> INFO:nlp.utils.info_utils:All the checksums matched successfully.
> INFO:nlp.builder:Generating split train
> INFO:nlp.arrow_writer:Done writing 285161 examples in 1240618482 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-train.arrow.
> INFO:nlp.builder:Generating split validation
> INFO:nlp.arrow_writer:Done writing 13255 examples in 56637485 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-validation.arrow.
> INFO:nlp.builder:Generating split test
> INFO:nlp.arrow_writer:Done writing 11379 examples in 48931393 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-test.arrow.
> Traceback (most recent call last):
> File "<stdin>", line 1, in <module>
> File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
> builder_instance.download_and_prepare(
> File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
> self._download_and_prepare(
> File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
> verify_splits(self.info.splits, split_dict)
> File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
> raise NonMatchingSplitsSizesError(str(bad_splits))
> nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
> ```
With `nlp == 0.3.0` version, I'm not able to reproduce this error on my side.
Which version are you using for reproducing your bug?
```
>> nlp.load_dataset('cnn_dailymail', '3.0.0')
8.90k/8.90k [00:18<00:00, 486B/s]
Downloading: 100%
9.37k/9.37k [00:00<00:00, 234kB/s]
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /root/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Downloading:
159M/? [00:09<00:00, 16.7MB/s]
Downloading:
376M/? [00:06<00:00, 62.6MB/s]
Downloading:
2.11M/? [00:06<00:00, 333kB/s]
Downloading:
46.4M/? [00:02<00:00, 18.4MB/s]
Downloading:
2.43M/? [00:00<00:00, 2.62MB/s]
Dataset cnn_dailymail downloaded and prepared to /root/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0. Subsequent calls will reuse this data.
{'test': Dataset(schema: {'article': 'string', 'highlights': 'string'}, num_rows: 11490),
'train': Dataset(schema: {'article': 'string', 'highlights': 'string'}, num_rows: 287113),
'validation': Dataset(schema: {'article': 'string', 'highlights': 'string'}, num_rows: 13368)}
>> ...
``` | ```
>>> import nlp
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
``` | 376 | Loading CNN/Daily Mail dataset produces `nlp.utils.info_utils.NonMatchingSplitsSizesError`
```
>>> import nlp
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
```
> here's the log
>
> ```
> >>> import nlp
> import logging
> logging.basicConfig(level=logging.INFO)
> nlp.load_dataset('cnn_dailymail', '3.0.0')
> >>> import logging
> >>> logging.basicConfig(level=logging.INFO)
> >>> nlp.load_dataset('cnn_dailymail', '3.0.0')
> INFO:nlp.load:Checking /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py for additional imports.
> INFO:filelock:Lock 140443095301136 acquired on /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py.lock
> INFO:nlp.load:Found main folder for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail
> INFO:nlp.load:Found specific version folder for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad
> INFO:nlp.load:Found script file from https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py to /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/cnn_dailymail.py
> INFO:nlp.load:Updating dataset infos file from https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/dataset_infos.json to /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/dataset_infos.json
> INFO:nlp.load:Found metadata file for dataset https://s3.amazonaws.com/datasets.huggingface.co/nlp/datasets/cnn_dailymail/cnn_dailymail.py at /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad/cnn_dailymail.json
> INFO:filelock:Lock 140443095301136 released on /u/jm8wx/.cache/huggingface/datasets/720d2e20d8dc6d98f21195a39cc934bb41dd0a40b57ea3d323661a7c5d70522c.d44c2417f4e0fe938ede0a684dcbb1fa9b4789de22e8a99c43103d4b4c374b3b.py.lock
> INFO:nlp.info:Loading Dataset Infos from /p/qdata/jm8wx/datasets/nlp/src/nlp/datasets/cnn_dailymail/9645e0bc96f647decf46541f6f4bef6936ee82ace653ac362bab03309a46d4ad
> INFO:nlp.builder:Generating dataset cnn_dailymail (/u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0)
> INFO:nlp.builder:Dataset not on Hf google storage. Downloading and preparing it from source
> Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
> INFO:nlp.utils.info_utils:All the checksums matched successfully.
> INFO:nlp.builder:Generating split train
> INFO:nlp.arrow_writer:Done writing 285161 examples in 1240618482 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-train.arrow.
> INFO:nlp.builder:Generating split validation
> INFO:nlp.arrow_writer:Done writing 13255 examples in 56637485 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-validation.arrow.
> INFO:nlp.builder:Generating split test
> INFO:nlp.arrow_writer:Done writing 11379 examples in 48931393 bytes /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0.incomplete/cnn_dailymail-test.arrow.
> Traceback (most recent call last):
> File "<stdin>", line 1, in <module>
> File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
> builder_instance.download_and_prepare(
> File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
> self._download_and_prepare(
> File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
> verify_splits(self.info.splits, split_dict)
> File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
> raise NonMatchingSplitsSizesError(str(bad_splits))
> nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
> ```
With `nlp == 0.3.0` version, I'm not able to reproduce this error on my side.
Which version are you using for reproducing your bug?
```
>> nlp.load_dataset('cnn_dailymail', '3.0.0')
8.90k/8.90k [00:18<00:00, 486B/s]
Downloading: 100%
9.37k/9.37k [00:00<00:00, 234kB/s]
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /root/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Downloading:
159M/? [00:09<00:00, 16.7MB/s]
Downloading:
376M/? [00:06<00:00, 62.6MB/s]
Downloading:
2.11M/? [00:06<00:00, 333kB/s]
Downloading:
46.4M/? [00:02<00:00, 18.4MB/s]
Downloading:
2.43M/? [00:00<00:00, 2.62MB/s]
Dataset cnn_dailymail downloaded and prepared to /root/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0. Subsequent calls will reuse this data.
{'test': Dataset(schema: {'article': 'string', 'highlights': 'string'}, num_rows: 11490),
'train': Dataset(schema: {'article': 'string', 'highlights': 'string'}, num_rows: 287113),
'validation': Dataset(schema: {'article': 'string', 'highlights': 'string'}, num_rows: 13368)}
>> ...
``` |
https://github.com/huggingface/datasets/issues/331 | Loading CNN/Daily Mail dataset produces `nlp.utils.info_utils.NonMatchingSplitsSizesError` | In general if some examples are missing after processing (hence causing the `NonMatchingSplitsSizesError `), it is often due to either
1) corrupted cached files
2) decoding errors
I just checked the dataset script for code that could lead to decoding errors but I couldn't find any. Before we try to dive more into the processing of the dataset, could you try to clear your cache ? Just to make sure that it isn't 1) | ```
>>> import nlp
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
``` | 74 | Loading CNN/Daily Mail dataset produces `nlp.utils.info_utils.NonMatchingSplitsSizesError`
```
>>> import nlp
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
```
In general if some examples are missing after processing (hence causing the `NonMatchingSplitsSizesError `), it is often due to either
1) corrupted cached files
2) decoding errors
I just checked the dataset script for code that could lead to decoding errors but I couldn't find any. Before we try to dive more into the processing of the dataset, could you try to clear your cache ? Just to make sure that it isn't 1) |
https://github.com/huggingface/datasets/issues/331 | Loading CNN/Daily Mail dataset produces `nlp.utils.info_utils.NonMatchingSplitsSizesError` | Yes thanks for the support! I cleared out my cache folder and everything works fine now | ```
>>> import nlp
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
``` | 16 | Loading CNN/Daily Mail dataset produces `nlp.utils.info_utils.NonMatchingSplitsSizesError`
```
>>> import nlp
>>> nlp.load_dataset('cnn_dailymail', '3.0.0')
Downloading and preparing dataset cnn_dailymail/3.0.0 (download: 558.32 MiB, generated: 1.26 GiB, total: 1.81 GiB) to /u/jm8wx/.cache/huggingface/datasets/cnn_dailymail/3.0.0/3.0.0...
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/load.py", line 520, in load_dataset
builder_instance.download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 431, in download_and_prepare
self._download_and_prepare(
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/p/qdata/jm8wx/datasets/nlp/src/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
nlp.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='test', num_bytes=49424491, num_examples=11490, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='test', num_bytes=48931393, num_examples=11379, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='train', num_bytes=1249178681, num_examples=287113, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='train', num_bytes=1240618482, num_examples=285161, dataset_name='cnn_dailymail')}, {'expected': SplitInfo(name='validation', num_bytes=57149241, num_examples=13368, dataset_name='cnn_dailymail'), 'recorded': SplitInfo(name='validation', num_bytes=56637485, num_examples=13255, dataset_name='cnn_dailymail')}]
```
Yes thanks for the support! I cleared out my cache folder and everything works fine now |
https://github.com/huggingface/datasets/issues/329 | [Bug] FileLock dependency incompatible with filesystem | Environment is Ubuntu 18.04, Python 3.7.5, nlp==0.3.0, filelock=3.0.12.
The external volume is Amazon FSx for Lustre, and it by default creates files with limited permissions. My working theory is that FileLock creates a lockfile that isn't writable, and thus there's no way to acquire it by removing the .lock file. But Python is able to create new files and write to them outside of the FileLock package.
When I attempt to use FileLock within a Docker container by writing to `/root/.cache/hello.txt`, it succeeds. So there's some permissions issue. But it's not a Docker configuration issue; I've replicated it without Docker.
```bash
echo "hello world" >> hello.txt
ls -l
-rw-rw-r-- 1 ubuntu ubuntu 10 Jun 30 19:52 hello.txt
``` | I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that. | 118 | [Bug] FileLock dependency incompatible with filesystem
I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that.
Environment is Ubuntu 18.04, Python 3.7.5, nlp==0.3.0, filelock=3.0.12.
The external volume is Amazon FSx for Lustre, and it by default creates files with limited permissions. My working theory is that FileLock creates a lockfile that isn't writable, and thus there's no way to acquire it by removing the .lock file. But Python is able to create new files and write to them outside of the FileLock package.
When I attempt to use FileLock within a Docker container by writing to `/root/.cache/hello.txt`, it succeeds. So there's some permissions issue. But it's not a Docker configuration issue; I've replicated it without Docker.
```bash
echo "hello world" >> hello.txt
ls -l
-rw-rw-r-- 1 ubuntu ubuntu 10 Jun 30 19:52 hello.txt
``` |
https://github.com/huggingface/datasets/issues/329 | [Bug] FileLock dependency incompatible with filesystem | Looks like the `flock` syscall does not work on Lustre filesystems by default: https://github.com/benediktschmitt/py-filelock/issues/67.
I added the `-o flock` option when mounting the filesystem, as [described here](https://docs.aws.amazon.com/fsx/latest/LustreGuide/getting-started-step2.html), which fixed the issue. | I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that. | 31 | [Bug] FileLock dependency incompatible with filesystem
I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that.
Looks like the `flock` syscall does not work on Lustre filesystems by default: https://github.com/benediktschmitt/py-filelock/issues/67.
I added the `-o flock` option when mounting the filesystem, as [described here](https://docs.aws.amazon.com/fsx/latest/LustreGuide/getting-started-step2.html), which fixed the issue. |
https://github.com/huggingface/datasets/issues/329 | [Bug] FileLock dependency incompatible with filesystem | I'm wondering if this can be revisited. In some managed environments the same person using HF cannot change the file-system mount flags, (and the organization may be unwilling to change these flags due to other concerns) but can ensure that there won't be concurrent writes, for example because HF is offline and the models/datasets were downloaded earlier.
The real fix would be to FileLock itself, which does not seem very active and seems to not deal with failed system flock calls , which would be one way to fix this, as they mention in the issue below also raised by @jarednielsen
https://github.com/tox-dev/py-filelock/issues/67 | I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that. | 102 | [Bug] FileLock dependency incompatible with filesystem
I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that.
I'm wondering if this can be revisited. In some managed environments the same person using HF cannot change the file-system mount flags, (and the organization may be unwilling to change these flags due to other concerns) but can ensure that there won't be concurrent writes, for example because HF is offline and the models/datasets were downloaded earlier.
The real fix would be to FileLock itself, which does not seem very active and seems to not deal with failed system flock calls , which would be one way to fix this, as they mention in the issue below also raised by @jarednielsen
https://github.com/tox-dev/py-filelock/issues/67 |
https://github.com/huggingface/datasets/issues/329 | [Bug] FileLock dependency incompatible with filesystem | > I'm wondering if this can be revisited. In some managed environments the same person using HF cannot change the file-system mount flags, (and the organization may be unwilling to change these flags due to other concerns) but can ensure that there won't be concurrent writes, for example because HF is offline and the models/datasets were downloaded earlier.
I am one of those users. Is there a work around for this?
| I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that. | 71 | [Bug] FileLock dependency incompatible with filesystem
I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that.
> I'm wondering if this can be revisited. In some managed environments the same person using HF cannot change the file-system mount flags, (and the organization may be unwilling to change these flags due to other concerns) but can ensure that there won't be concurrent writes, for example because HF is offline and the models/datasets were downloaded earlier.
I am one of those users. Is there a work around for this?
|
https://github.com/huggingface/datasets/issues/329 | [Bug] FileLock dependency incompatible with filesystem | The machines I use have a shared FS which has the filelock problem as well as a local one that does not. Using some env vars (HF_HOME, which controls both models and datasets, and HF_DATASETS_OFFLINE) for both transformers and datasets library one can influence where these downloads happen, and whether the locks get taken. I think some of the relevant documentation is here https://huggingface.co/docs/transformers/installation#cache-setup. I do end up using different settings when I download the models and when I use them, and have to rsync the models to the local file system using a separate script. | I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that. | 96 | [Bug] FileLock dependency incompatible with filesystem
I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that.
The machines I use have a shared FS which has the filelock problem as well as a local one that does not. Using some env vars (HF_HOME, which controls both models and datasets, and HF_DATASETS_OFFLINE) for both transformers and datasets library one can influence where these downloads happen, and whether the locks get taken. I think some of the relevant documentation is here https://huggingface.co/docs/transformers/installation#cache-setup. I do end up using different settings when I download the models and when I use them, and have to rsync the models to the local file system using a separate script. |
https://github.com/huggingface/datasets/issues/329 | [Bug] FileLock dependency incompatible with filesystem | Thanks @orm011 . These filesystems are such a pain. I'll dig around, looks like setting `cache_dir` to a non-lustre filesystem works for `transformers` but not `datasets`. | I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that. | 26 | [Bug] FileLock dependency incompatible with filesystem
I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that.
Thanks @orm011 . These filesystems are such a pain. I'll dig around, looks like setting `cache_dir` to a non-lustre filesystem works for `transformers` but not `datasets`. |
https://github.com/huggingface/datasets/issues/329 | [Bug] FileLock dependency incompatible with filesystem | Note I `export HF_HOME=` in the shell prior to running python (I do not use the `cache_dir` argument, I think I ran into similar issues with it, nor `HF_DATASETS_CACHE` , though maybe that works, or maybe you can set it in python prior to importing the library ), and I change no other variables. Then `datasets.load_dataset()` works without any additional flags, and they go into `HF_HOME/datasets/` and the models go into `HF_HOME/transformers/` (and the lock files are all there as well). | I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that. | 81 | [Bug] FileLock dependency incompatible with filesystem
I'm downloading a dataset successfully with
`load_dataset("wikitext", "wikitext-2-raw-v1")`
But when I attempt to cache it on an external volume, it hangs indefinitely:
`load_dataset("wikitext", "wikitext-2-raw-v1", cache_dir="/fsx") # /fsx is an external volume mount`
The filesystem when hanging looks like this:
```bash
/fsx
----downloads
----94be...73.lock
----wikitext
----wikitext-2-raw
----wikitext-2-raw-1.0.0.incomplete
```
It appears that on this filesystem, the FileLock object is forever stuck in its "acquire" stage. I have verified that the issue lies specifically with the `filelock` dependency:
```python
open("/fsx/hello.txt").write("hello") # succeeds
from filelock import FileLock
with FileLock("/fsx/hello.lock"):
open("/fsx/hello.txt").write("hello") # hangs indefinitely
```
Has anyone else run into this issue? I'd raise it directly on the FileLock repo, but that project appears abandoned with the last update over a year ago. Or if there's a solution that would remove the FileLock dependency from the project, I would appreciate that.
Note I `export HF_HOME=` in the shell prior to running python (I do not use the `cache_dir` argument, I think I ran into similar issues with it, nor `HF_DATASETS_CACHE` , though maybe that works, or maybe you can set it in python prior to importing the library ), and I change no other variables. Then `datasets.load_dataset()` works without any additional flags, and they go into `HF_HOME/datasets/` and the models go into `HF_HOME/transformers/` (and the lock files are all there as well). |
https://github.com/huggingface/datasets/issues/328 | Fork dataset | To be able to generate the Arrow dataset you need to either use our csv or json utilities `load_dataset("json", data_files=my_json_files)` OR write your own custom dataset script (you can find some inspiration from the [squad](https://github.com/huggingface/nlp/blob/master/datasets/squad/squad.py) script for example). Custom dataset scripts can be called locally with `nlp.load_dataset(path_to_my_script_directory)`.
This should help you get what you call "Dataset1".
Then using some dataset transforms like `.map` for example you can get to "DatasetNER" and "DatasetREL".
| We have a multi-task learning model training I'm trying to convert to using the Arrow-based nlp dataset.
We're currently training a custom TensorFlow model but the nlp paradigm should be a bridge for us to be able to use the wealth of pre-trained models in Transformers.
Our preprocessing flow parses raw text and json with Entity and Relations annotations and creates 2 datasets for training a NER and Relations prediction heads.
Is there some good way to "fork" dataset-
EG
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 -> DatasetREL
or
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 + DatasetNER -> DatasetREL
| 72 | Fork dataset
We have a multi-task learning model training I'm trying to convert to using the Arrow-based nlp dataset.
We're currently training a custom TensorFlow model but the nlp paradigm should be a bridge for us to be able to use the wealth of pre-trained models in Transformers.
Our preprocessing flow parses raw text and json with Entity and Relations annotations and creates 2 datasets for training a NER and Relations prediction heads.
Is there some good way to "fork" dataset-
EG
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 -> DatasetREL
or
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 + DatasetNER -> DatasetREL
To be able to generate the Arrow dataset you need to either use our csv or json utilities `load_dataset("json", data_files=my_json_files)` OR write your own custom dataset script (you can find some inspiration from the [squad](https://github.com/huggingface/nlp/blob/master/datasets/squad/squad.py) script for example). Custom dataset scripts can be called locally with `nlp.load_dataset(path_to_my_script_directory)`.
This should help you get what you call "Dataset1".
Then using some dataset transforms like `.map` for example you can get to "DatasetNER" and "DatasetREL".
|
https://github.com/huggingface/datasets/issues/328 | Fork dataset | Thanks for the helpful advice, @lhoestq -- I wasn't quite able to get the json recipe working -
```
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/pyarrow/ipc.py in __init__(self, source)
60
61 def __init__(self, source):
---> 62 self._open(source)
63
64
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/pyarrow/ipc.pxi in pyarrow.lib._RecordBatchStreamReader._open()
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/pyarrow/error.pxi in pyarrow.lib.pyarrow_internal_check_status()
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/pyarrow/error.pxi in pyarrow.lib.check_status()
ArrowInvalid: Tried reading schema message, was null or length 0
```
But I'm going to give the generator_dataset_builder a try.
1 more quick question -- can .map be used to output different length mappings -- could I skip one, or yield 2, can you map_batch | We have a multi-task learning model training I'm trying to convert to using the Arrow-based nlp dataset.
We're currently training a custom TensorFlow model but the nlp paradigm should be a bridge for us to be able to use the wealth of pre-trained models in Transformers.
Our preprocessing flow parses raw text and json with Entity and Relations annotations and creates 2 datasets for training a NER and Relations prediction heads.
Is there some good way to "fork" dataset-
EG
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 -> DatasetREL
or
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 + DatasetNER -> DatasetREL
| 87 | Fork dataset
We have a multi-task learning model training I'm trying to convert to using the Arrow-based nlp dataset.
We're currently training a custom TensorFlow model but the nlp paradigm should be a bridge for us to be able to use the wealth of pre-trained models in Transformers.
Our preprocessing flow parses raw text and json with Entity and Relations annotations and creates 2 datasets for training a NER and Relations prediction heads.
Is there some good way to "fork" dataset-
EG
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 -> DatasetREL
or
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 + DatasetNER -> DatasetREL
Thanks for the helpful advice, @lhoestq -- I wasn't quite able to get the json recipe working -
```
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/pyarrow/ipc.py in __init__(self, source)
60
61 def __init__(self, source):
---> 62 self._open(source)
63
64
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/pyarrow/ipc.pxi in pyarrow.lib._RecordBatchStreamReader._open()
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/pyarrow/error.pxi in pyarrow.lib.pyarrow_internal_check_status()
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/pyarrow/error.pxi in pyarrow.lib.check_status()
ArrowInvalid: Tried reading schema message, was null or length 0
```
But I'm going to give the generator_dataset_builder a try.
1 more quick question -- can .map be used to output different length mappings -- could I skip one, or yield 2, can you map_batch |
https://github.com/huggingface/datasets/issues/328 | Fork dataset | Thanks this answers my question. I think the issue I was having using the json loader were due to using gzipped jsonl files.
The error I get now is :
```
Using custom data configuration test
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-38-29082a31e5b2> in <module>
5 print(ner_datafiles)
6
----> 7 ds = nlp.load_dataset("json", "test", data_files=ner_datafiles[0])
8
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
736 schema_dict[field.name] = Value(str(field.type))
737
--> 738 parse_schema(writer.schema, features)
739 self.info.features = Features(features)
740
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in parse_schema(schema, schema_dict)
734 parse_schema(field.type.value_type, schema_dict[field.name])
735 else:
--> 736 schema_dict[field.name] = Value(str(field.type))
737
738 parse_schema(writer.schema, features)
<string> in __init__(self, dtype, id, _type)
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in __post_init__(self)
55
56 def __post_init__(self):
---> 57 self.pa_type = string_to_arrow(self.dtype)
58
59 def __call__(self):
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in string_to_arrow(type_str)
32 if str(type_str + "_") not in pa.__dict__:
33 raise ValueError(
---> 34 f"Neither {type_str} nor {type_str + '_'} seems to be a pyarrow data type. "
35 f"Please make sure to use a correct data type, see: "
36 f"https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions"
ValueError: Neither list<item: int64> nor list<item: int64>_ seems to be a pyarrow data type. Please make sure to use a correct data type, see: https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions.
```
If I just create a pa- table manually like is done in the jsonloader -- it seems to work fine. Ths JSON I'm trying to load isn't overly complex - 1 integer field, the rest text fields with a nested list of objects with text fields . | We have a multi-task learning model training I'm trying to convert to using the Arrow-based nlp dataset.
We're currently training a custom TensorFlow model but the nlp paradigm should be a bridge for us to be able to use the wealth of pre-trained models in Transformers.
Our preprocessing flow parses raw text and json with Entity and Relations annotations and creates 2 datasets for training a NER and Relations prediction heads.
Is there some good way to "fork" dataset-
EG
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 -> DatasetREL
or
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 + DatasetNER -> DatasetREL
| 324 | Fork dataset
We have a multi-task learning model training I'm trying to convert to using the Arrow-based nlp dataset.
We're currently training a custom TensorFlow model but the nlp paradigm should be a bridge for us to be able to use the wealth of pre-trained models in Transformers.
Our preprocessing flow parses raw text and json with Entity and Relations annotations and creates 2 datasets for training a NER and Relations prediction heads.
Is there some good way to "fork" dataset-
EG
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 -> DatasetREL
or
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 + DatasetNER -> DatasetREL
Thanks this answers my question. I think the issue I was having using the json loader were due to using gzipped jsonl files.
The error I get now is :
```
Using custom data configuration test
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-38-29082a31e5b2> in <module>
5 print(ner_datafiles)
6
----> 7 ds = nlp.load_dataset("json", "test", data_files=ner_datafiles[0])
8
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
736 schema_dict[field.name] = Value(str(field.type))
737
--> 738 parse_schema(writer.schema, features)
739 self.info.features = Features(features)
740
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/builder.py in parse_schema(schema, schema_dict)
734 parse_schema(field.type.value_type, schema_dict[field.name])
735 else:
--> 736 schema_dict[field.name] = Value(str(field.type))
737
738 parse_schema(writer.schema, features)
<string> in __init__(self, dtype, id, _type)
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in __post_init__(self)
55
56 def __post_init__(self):
---> 57 self.pa_type = string_to_arrow(self.dtype)
58
59 def __call__(self):
~/.virtualenvs/inv-text2struct/lib/python3.6/site-packages/nlp/features.py in string_to_arrow(type_str)
32 if str(type_str + "_") not in pa.__dict__:
33 raise ValueError(
---> 34 f"Neither {type_str} nor {type_str + '_'} seems to be a pyarrow data type. "
35 f"Please make sure to use a correct data type, see: "
36 f"https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions"
ValueError: Neither list<item: int64> nor list<item: int64>_ seems to be a pyarrow data type. Please make sure to use a correct data type, see: https://arrow.apache.org/docs/python/api/datatypes.html#factory-functions.
```
If I just create a pa- table manually like is done in the jsonloader -- it seems to work fine. Ths JSON I'm trying to load isn't overly complex - 1 integer field, the rest text fields with a nested list of objects with text fields . |
https://github.com/huggingface/datasets/issues/328 | Fork dataset | I'll close this -- It's still unclear how to go about troubleshooting the json example as I mentioned above. If I decide it's worth the trouble, I'll create another issue, or wait for a better support for using nlp for making custom data-loaders. | We have a multi-task learning model training I'm trying to convert to using the Arrow-based nlp dataset.
We're currently training a custom TensorFlow model but the nlp paradigm should be a bridge for us to be able to use the wealth of pre-trained models in Transformers.
Our preprocessing flow parses raw text and json with Entity and Relations annotations and creates 2 datasets for training a NER and Relations prediction heads.
Is there some good way to "fork" dataset-
EG
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 -> DatasetREL
or
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 + DatasetNER -> DatasetREL
| 43 | Fork dataset
We have a multi-task learning model training I'm trying to convert to using the Arrow-based nlp dataset.
We're currently training a custom TensorFlow model but the nlp paradigm should be a bridge for us to be able to use the wealth of pre-trained models in Transformers.
Our preprocessing flow parses raw text and json with Entity and Relations annotations and creates 2 datasets for training a NER and Relations prediction heads.
Is there some good way to "fork" dataset-
EG
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 -> DatasetREL
or
1. text + json -> Dataset1
1. Dataset1 -> DatasetNER
1. Dataset1 + DatasetNER -> DatasetREL
I'll close this -- It's still unclear how to go about troubleshooting the json example as I mentioned above. If I decide it's worth the trouble, I'll create another issue, or wait for a better support for using nlp for making custom data-loaders. |
https://github.com/huggingface/datasets/issues/326 | Large dataset in Squad2-format | I'm pretty sure you can get some inspiration from the squad_v2 script. It looks like the dataset is quite big so it will take some time for the users to generate it, but it should be reasonable.
Also you are saying that you are still making the dataset grow in size right ?
It's probably good practice to let the users do their training/evaluations with the exact same version of the dataset.
We allow for each dataset to specify a version (ex: 1.0.0) and increment this number every time there are new samples in the dataset for example. Does it look like a good solution for you ? Or would you rather have one final version with the full dataset ? | At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ? | 121 | Large dataset in Squad2-format
At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ?
I'm pretty sure you can get some inspiration from the squad_v2 script. It looks like the dataset is quite big so it will take some time for the users to generate it, but it should be reasonable.
Also you are saying that you are still making the dataset grow in size right ?
It's probably good practice to let the users do their training/evaluations with the exact same version of the dataset.
We allow for each dataset to specify a version (ex: 1.0.0) and increment this number every time there are new samples in the dataset for example. Does it look like a good solution for you ? Or would you rather have one final version with the full dataset ? |
https://github.com/huggingface/datasets/issues/326 | Large dataset in Squad2-format | It would also be good if there is any possibility for versioning, I think this way is much better than the dynamic way.
If you mean that part to put the tiles into one is the generation it would take up to 15-20 minutes on home computer hardware.
Are there any compression or optimization algorithms while generating the dataset ?
Otherwise the hardware limit is around 32 GB ram at the moment.
If everything works well we will add some more gigabytes of data in future what would make it pretty memory costly. | At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ? | 93 | Large dataset in Squad2-format
At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ?
It would also be good if there is any possibility for versioning, I think this way is much better than the dynamic way.
If you mean that part to put the tiles into one is the generation it would take up to 15-20 minutes on home computer hardware.
Are there any compression or optimization algorithms while generating the dataset ?
Otherwise the hardware limit is around 32 GB ram at the moment.
If everything works well we will add some more gigabytes of data in future what would make it pretty memory costly. |
https://github.com/huggingface/datasets/issues/326 | Large dataset in Squad2-format | 15-20 minutes is fine !
Also there's no RAM limitations as we save to disk every 1000 elements while generating the dataset by default.
After generation, the dataset is ready to use with (again) no RAM limitations as we do memory-mapping. | At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ? | 41 | Large dataset in Squad2-format
At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ?
15-20 minutes is fine !
Also there's no RAM limitations as we save to disk every 1000 elements while generating the dataset by default.
After generation, the dataset is ready to use with (again) no RAM limitations as we do memory-mapping. |
https://github.com/huggingface/datasets/issues/326 | Large dataset in Squad2-format | Wow, that sounds pretty cool.
Actually I have the problem of running out of memory while tokenization on our local machine.
That wouldn't happen again, would it ? | At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ? | 28 | Large dataset in Squad2-format
At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ?
Wow, that sounds pretty cool.
Actually I have the problem of running out of memory while tokenization on our local machine.
That wouldn't happen again, would it ? |
https://github.com/huggingface/datasets/issues/326 | Large dataset in Squad2-format | You can do the tokenization step using `my_tokenized_dataset = my_dataset.map(my_tokenize_function)` that writes the tokenized texts on disk as well. And then `my_tokenized_dataset` will be a memory-mapped dataset too, so you should be fine :) | At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ? | 34 | Large dataset in Squad2-format
At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ?
You can do the tokenization step using `my_tokenized_dataset = my_dataset.map(my_tokenize_function)` that writes the tokenized texts on disk as well. And then `my_tokenized_dataset` will be a memory-mapped dataset too, so you should be fine :) |
https://github.com/huggingface/datasets/issues/326 | Large dataset in Squad2-format | In your training loop, loading the tokenized texts is going to be fast and pretty much negligible compared to a forward pass. You shouldn't expect any slow down. | At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ? | 28 | Large dataset in Squad2-format
At the moment we are building an large question answering dataset and think about sharing it with the huggingface community.
Caused the computing power we splitted it into multiple tiles, but they are all in the same format.
Right now the most important facts about are this:
- Contexts: 1.047.671
- questions: 1.677.732
- Answers: 6.742.406
- unanswerable: 377.398
It is already cleaned
<pre><code>
train_data = [
{
'context': "this is the context",
'qas': [
{
'id': "00002",
'is_impossible': False,
'question': "whats is this",
'answers': [
{
'text': "answer",
'answer_start': 0
}
]
},
{
'id': "00003",
'is_impossible': False,
'question': "question2",
'answers': [
{
'text': "answer2",
'answer_start': 1
}
]
}
]
}
]
</code></pre>
Cause it is growing every day we are thinking about an structure like this:
We host an Json file, containing all the download links and the script can load it dynamically.
At the moment it is around ~20GB
Any advice how to handle this, or an ready to use template ?
In your training loop, loading the tokenized texts is going to be fast and pretty much negligible compared to a forward pass. You shouldn't expect any slow down. |
https://github.com/huggingface/datasets/issues/324 | Error when calculating glue score | The glue metric for cola is a metric for classification. It expects label ids as integers as inputs. | I was trying glue score along with other metrics here. But glue gives me this error;
```
import nlp
glue_metric = nlp.load_metric('glue',name="cola")
glue_score = glue_metric.compute(predictions, references)
```
```
---------------------------------------------------------------------------
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-8-b9210a524504> in <module>()
----> 1 glue_score = glue_metric.compute(predictions, references)
6 frames
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in compute(self, predictions, references, timeout, **metrics_kwargs)
191 """
192 if predictions is not None:
--> 193 self.add_batch(predictions=predictions, references=references)
194 self.finalize(timeout=timeout)
195
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in add_batch(self, predictions, references, **kwargs)
207 if self.writer is None:
208 self._init_writer()
--> 209 self.writer.write_batch(batch)
210
211 def add(self, prediction=None, reference=None, **kwargs):
/usr/local/lib/python3.6/dist-packages/nlp/arrow_writer.py in write_batch(self, batch_examples, writer_batch_size)
155 if self.pa_writer is None:
156 self._build_writer(pa_table=pa.Table.from_pydict(batch_examples))
--> 157 pa_table: pa.Table = pa.Table.from_pydict(batch_examples, schema=self._schema)
158 if writer_batch_size is None:
159 writer_batch_size = self.writer_batch_size
/usr/local/lib/python3.6/dist-packages/pyarrow/types.pxi in __iter__()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.asarray()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.array()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib._sequence_to_array()
TypeError: an integer is required (got type str)
```
I'm not sure whether I'm doing this wrong or whether it's an issue. I would like to know a workaround. Thank you. | 18 | Error when calculating glue score
I was trying glue score along with other metrics here. But glue gives me this error;
```
import nlp
glue_metric = nlp.load_metric('glue',name="cola")
glue_score = glue_metric.compute(predictions, references)
```
```
---------------------------------------------------------------------------
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-8-b9210a524504> in <module>()
----> 1 glue_score = glue_metric.compute(predictions, references)
6 frames
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in compute(self, predictions, references, timeout, **metrics_kwargs)
191 """
192 if predictions is not None:
--> 193 self.add_batch(predictions=predictions, references=references)
194 self.finalize(timeout=timeout)
195
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in add_batch(self, predictions, references, **kwargs)
207 if self.writer is None:
208 self._init_writer()
--> 209 self.writer.write_batch(batch)
210
211 def add(self, prediction=None, reference=None, **kwargs):
/usr/local/lib/python3.6/dist-packages/nlp/arrow_writer.py in write_batch(self, batch_examples, writer_batch_size)
155 if self.pa_writer is None:
156 self._build_writer(pa_table=pa.Table.from_pydict(batch_examples))
--> 157 pa_table: pa.Table = pa.Table.from_pydict(batch_examples, schema=self._schema)
158 if writer_batch_size is None:
159 writer_batch_size = self.writer_batch_size
/usr/local/lib/python3.6/dist-packages/pyarrow/types.pxi in __iter__()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.asarray()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.array()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib._sequence_to_array()
TypeError: an integer is required (got type str)
```
I'm not sure whether I'm doing this wrong or whether it's an issue. I would like to know a workaround. Thank you.
The glue metric for cola is a metric for classification. It expects label ids as integers as inputs. |
https://github.com/huggingface/datasets/issues/324 | Error when calculating glue score | I want to evaluate a sentence pair whether they are semantically equivalent, so I used MRPC and it gives the same error, does that mean we have to encode the sentences and parse as input?
using BertTokenizer;
```
encoded_reference=tokenizer.encode(reference, add_special_tokens=False)
encoded_prediction=tokenizer.encode(prediction, add_special_tokens=False)
```
`glue_score = glue_metric.compute(encoded_prediction, encoded_reference)`
```
ValueError Traceback (most recent call last)
<ipython-input-9-4c3a3ce7b583> in <module>()
----> 1 glue_score = glue_metric.compute(encoded_prediction, encoded_reference)
6 frames
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in compute(self, predictions, references, timeout, **metrics_kwargs)
198 predictions = self.data["predictions"]
199 references = self.data["references"]
--> 200 output = self._compute(predictions=predictions, references=references, **metrics_kwargs)
201 return output
202
/usr/local/lib/python3.6/dist-packages/nlp/metrics/glue/27b1bc63e520833054bd0d7a8d0bc7f6aab84cc9eed1b576e98c806f9466d302/glue.py in _compute(self, predictions, references)
101 return pearson_and_spearman(predictions, references)
102 elif self.config_name in ["mrpc", "qqp"]:
--> 103 return acc_and_f1(predictions, references)
104 elif self.config_name in ["sst2", "mnli", "mnli_mismatched", "mnli_matched", "qnli", "rte", "wnli", "hans"]:
105 return {"accuracy": simple_accuracy(predictions, references)}
/usr/local/lib/python3.6/dist-packages/nlp/metrics/glue/27b1bc63e520833054bd0d7a8d0bc7f6aab84cc9eed1b576e98c806f9466d302/glue.py in acc_and_f1(preds, labels)
60 def acc_and_f1(preds, labels):
61 acc = simple_accuracy(preds, labels)
---> 62 f1 = f1_score(y_true=labels, y_pred=preds)
63 return {
64 "accuracy": acc,
/usr/local/lib/python3.6/dist-packages/sklearn/metrics/_classification.py in f1_score(y_true, y_pred, labels, pos_label, average, sample_weight, zero_division)
1097 pos_label=pos_label, average=average,
1098 sample_weight=sample_weight,
-> 1099 zero_division=zero_division)
1100
1101
/usr/local/lib/python3.6/dist-packages/sklearn/metrics/_classification.py in fbeta_score(y_true, y_pred, beta, labels, pos_label, average, sample_weight, zero_division)
1224 warn_for=('f-score',),
1225 sample_weight=sample_weight,
-> 1226 zero_division=zero_division)
1227 return f
1228
/usr/local/lib/python3.6/dist-packages/sklearn/metrics/_classification.py in precision_recall_fscore_support(y_true, y_pred, beta, labels, pos_label, average, warn_for, sample_weight, zero_division)
1482 raise ValueError("beta should be >=0 in the F-beta score")
1483 labels = _check_set_wise_labels(y_true, y_pred, average, labels,
-> 1484 pos_label)
1485
1486 # Calculate tp_sum, pred_sum, true_sum ###
/usr/local/lib/python3.6/dist-packages/sklearn/metrics/_classification.py in _check_set_wise_labels(y_true, y_pred, average, labels, pos_label)
1314 raise ValueError("Target is %s but average='binary'. Please "
1315 "choose another average setting, one of %r."
-> 1316 % (y_type, average_options))
1317 elif pos_label not in (None, 1):
1318 warnings.warn("Note that pos_label (set to %r) is ignored when "
ValueError: Target is multiclass but average='binary'. Please choose another average setting, one of [None, 'micro', 'macro', 'weighted'].
``` | I was trying glue score along with other metrics here. But glue gives me this error;
```
import nlp
glue_metric = nlp.load_metric('glue',name="cola")
glue_score = glue_metric.compute(predictions, references)
```
```
---------------------------------------------------------------------------
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-8-b9210a524504> in <module>()
----> 1 glue_score = glue_metric.compute(predictions, references)
6 frames
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in compute(self, predictions, references, timeout, **metrics_kwargs)
191 """
192 if predictions is not None:
--> 193 self.add_batch(predictions=predictions, references=references)
194 self.finalize(timeout=timeout)
195
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in add_batch(self, predictions, references, **kwargs)
207 if self.writer is None:
208 self._init_writer()
--> 209 self.writer.write_batch(batch)
210
211 def add(self, prediction=None, reference=None, **kwargs):
/usr/local/lib/python3.6/dist-packages/nlp/arrow_writer.py in write_batch(self, batch_examples, writer_batch_size)
155 if self.pa_writer is None:
156 self._build_writer(pa_table=pa.Table.from_pydict(batch_examples))
--> 157 pa_table: pa.Table = pa.Table.from_pydict(batch_examples, schema=self._schema)
158 if writer_batch_size is None:
159 writer_batch_size = self.writer_batch_size
/usr/local/lib/python3.6/dist-packages/pyarrow/types.pxi in __iter__()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.asarray()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.array()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib._sequence_to_array()
TypeError: an integer is required (got type str)
```
I'm not sure whether I'm doing this wrong or whether it's an issue. I would like to know a workaround. Thank you. | 297 | Error when calculating glue score
I was trying glue score along with other metrics here. But glue gives me this error;
```
import nlp
glue_metric = nlp.load_metric('glue',name="cola")
glue_score = glue_metric.compute(predictions, references)
```
```
---------------------------------------------------------------------------
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-8-b9210a524504> in <module>()
----> 1 glue_score = glue_metric.compute(predictions, references)
6 frames
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in compute(self, predictions, references, timeout, **metrics_kwargs)
191 """
192 if predictions is not None:
--> 193 self.add_batch(predictions=predictions, references=references)
194 self.finalize(timeout=timeout)
195
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in add_batch(self, predictions, references, **kwargs)
207 if self.writer is None:
208 self._init_writer()
--> 209 self.writer.write_batch(batch)
210
211 def add(self, prediction=None, reference=None, **kwargs):
/usr/local/lib/python3.6/dist-packages/nlp/arrow_writer.py in write_batch(self, batch_examples, writer_batch_size)
155 if self.pa_writer is None:
156 self._build_writer(pa_table=pa.Table.from_pydict(batch_examples))
--> 157 pa_table: pa.Table = pa.Table.from_pydict(batch_examples, schema=self._schema)
158 if writer_batch_size is None:
159 writer_batch_size = self.writer_batch_size
/usr/local/lib/python3.6/dist-packages/pyarrow/types.pxi in __iter__()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.asarray()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.array()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib._sequence_to_array()
TypeError: an integer is required (got type str)
```
I'm not sure whether I'm doing this wrong or whether it's an issue. I would like to know a workaround. Thank you.
I want to evaluate a sentence pair whether they are semantically equivalent, so I used MRPC and it gives the same error, does that mean we have to encode the sentences and parse as input?
using BertTokenizer;
```
encoded_reference=tokenizer.encode(reference, add_special_tokens=False)
encoded_prediction=tokenizer.encode(prediction, add_special_tokens=False)
```
`glue_score = glue_metric.compute(encoded_prediction, encoded_reference)`
```
ValueError Traceback (most recent call last)
<ipython-input-9-4c3a3ce7b583> in <module>()
----> 1 glue_score = glue_metric.compute(encoded_prediction, encoded_reference)
6 frames
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in compute(self, predictions, references, timeout, **metrics_kwargs)
198 predictions = self.data["predictions"]
199 references = self.data["references"]
--> 200 output = self._compute(predictions=predictions, references=references, **metrics_kwargs)
201 return output
202
/usr/local/lib/python3.6/dist-packages/nlp/metrics/glue/27b1bc63e520833054bd0d7a8d0bc7f6aab84cc9eed1b576e98c806f9466d302/glue.py in _compute(self, predictions, references)
101 return pearson_and_spearman(predictions, references)
102 elif self.config_name in ["mrpc", "qqp"]:
--> 103 return acc_and_f1(predictions, references)
104 elif self.config_name in ["sst2", "mnli", "mnli_mismatched", "mnli_matched", "qnli", "rte", "wnli", "hans"]:
105 return {"accuracy": simple_accuracy(predictions, references)}
/usr/local/lib/python3.6/dist-packages/nlp/metrics/glue/27b1bc63e520833054bd0d7a8d0bc7f6aab84cc9eed1b576e98c806f9466d302/glue.py in acc_and_f1(preds, labels)
60 def acc_and_f1(preds, labels):
61 acc = simple_accuracy(preds, labels)
---> 62 f1 = f1_score(y_true=labels, y_pred=preds)
63 return {
64 "accuracy": acc,
/usr/local/lib/python3.6/dist-packages/sklearn/metrics/_classification.py in f1_score(y_true, y_pred, labels, pos_label, average, sample_weight, zero_division)
1097 pos_label=pos_label, average=average,
1098 sample_weight=sample_weight,
-> 1099 zero_division=zero_division)
1100
1101
/usr/local/lib/python3.6/dist-packages/sklearn/metrics/_classification.py in fbeta_score(y_true, y_pred, beta, labels, pos_label, average, sample_weight, zero_division)
1224 warn_for=('f-score',),
1225 sample_weight=sample_weight,
-> 1226 zero_division=zero_division)
1227 return f
1228
/usr/local/lib/python3.6/dist-packages/sklearn/metrics/_classification.py in precision_recall_fscore_support(y_true, y_pred, beta, labels, pos_label, average, warn_for, sample_weight, zero_division)
1482 raise ValueError("beta should be >=0 in the F-beta score")
1483 labels = _check_set_wise_labels(y_true, y_pred, average, labels,
-> 1484 pos_label)
1485
1486 # Calculate tp_sum, pred_sum, true_sum ###
/usr/local/lib/python3.6/dist-packages/sklearn/metrics/_classification.py in _check_set_wise_labels(y_true, y_pred, average, labels, pos_label)
1314 raise ValueError("Target is %s but average='binary'. Please "
1315 "choose another average setting, one of %r."
-> 1316 % (y_type, average_options))
1317 elif pos_label not in (None, 1):
1318 warnings.warn("Note that pos_label (set to %r) is ignored when "
ValueError: Target is multiclass but average='binary'. Please choose another average setting, one of [None, 'micro', 'macro', 'weighted'].
``` |
https://github.com/huggingface/datasets/issues/324 | Error when calculating glue score | MRPC is also a binary classification task, so its metric is a binary classification metric.
To evaluate if pairs of sentences are semantically equivalent, maybe you could take a look at models that compute if one sentence entails the other or not (typically the kinds of model that could work well on the MRPC task). | I was trying glue score along with other metrics here. But glue gives me this error;
```
import nlp
glue_metric = nlp.load_metric('glue',name="cola")
glue_score = glue_metric.compute(predictions, references)
```
```
---------------------------------------------------------------------------
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-8-b9210a524504> in <module>()
----> 1 glue_score = glue_metric.compute(predictions, references)
6 frames
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in compute(self, predictions, references, timeout, **metrics_kwargs)
191 """
192 if predictions is not None:
--> 193 self.add_batch(predictions=predictions, references=references)
194 self.finalize(timeout=timeout)
195
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in add_batch(self, predictions, references, **kwargs)
207 if self.writer is None:
208 self._init_writer()
--> 209 self.writer.write_batch(batch)
210
211 def add(self, prediction=None, reference=None, **kwargs):
/usr/local/lib/python3.6/dist-packages/nlp/arrow_writer.py in write_batch(self, batch_examples, writer_batch_size)
155 if self.pa_writer is None:
156 self._build_writer(pa_table=pa.Table.from_pydict(batch_examples))
--> 157 pa_table: pa.Table = pa.Table.from_pydict(batch_examples, schema=self._schema)
158 if writer_batch_size is None:
159 writer_batch_size = self.writer_batch_size
/usr/local/lib/python3.6/dist-packages/pyarrow/types.pxi in __iter__()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.asarray()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.array()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib._sequence_to_array()
TypeError: an integer is required (got type str)
```
I'm not sure whether I'm doing this wrong or whether it's an issue. I would like to know a workaround. Thank you. | 55 | Error when calculating glue score
I was trying glue score along with other metrics here. But glue gives me this error;
```
import nlp
glue_metric = nlp.load_metric('glue',name="cola")
glue_score = glue_metric.compute(predictions, references)
```
```
---------------------------------------------------------------------------
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-8-b9210a524504> in <module>()
----> 1 glue_score = glue_metric.compute(predictions, references)
6 frames
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in compute(self, predictions, references, timeout, **metrics_kwargs)
191 """
192 if predictions is not None:
--> 193 self.add_batch(predictions=predictions, references=references)
194 self.finalize(timeout=timeout)
195
/usr/local/lib/python3.6/dist-packages/nlp/metric.py in add_batch(self, predictions, references, **kwargs)
207 if self.writer is None:
208 self._init_writer()
--> 209 self.writer.write_batch(batch)
210
211 def add(self, prediction=None, reference=None, **kwargs):
/usr/local/lib/python3.6/dist-packages/nlp/arrow_writer.py in write_batch(self, batch_examples, writer_batch_size)
155 if self.pa_writer is None:
156 self._build_writer(pa_table=pa.Table.from_pydict(batch_examples))
--> 157 pa_table: pa.Table = pa.Table.from_pydict(batch_examples, schema=self._schema)
158 if writer_batch_size is None:
159 writer_batch_size = self.writer_batch_size
/usr/local/lib/python3.6/dist-packages/pyarrow/types.pxi in __iter__()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.asarray()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib.array()
/usr/local/lib/python3.6/dist-packages/pyarrow/array.pxi in pyarrow.lib._sequence_to_array()
TypeError: an integer is required (got type str)
```
I'm not sure whether I'm doing this wrong or whether it's an issue. I would like to know a workaround. Thank you.
MRPC is also a binary classification task, so its metric is a binary classification metric.
To evaluate if pairs of sentences are semantically equivalent, maybe you could take a look at models that compute if one sentence entails the other or not (typically the kinds of model that could work well on the MRPC task). |
https://github.com/huggingface/datasets/issues/321 | ERROR:root:mwparserfromhell | It looks like it comes from `mwparserfromhell`.
Would it be possible to get the bad `section` that causes this issue ? The `section` string is from `datasets/wikipedia.py:L548` ? You could just add a `try` statement and print the section if the line `section_text.append(section.strip_code().strip())` crashes.
It will help us know if we have to fix it on our side or if it is a `mwparserfromhell` issue. | Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
| 65 | ERROR:root:mwparserfromhell
Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
It looks like it comes from `mwparserfromhell`.
Would it be possible to get the bad `section` that causes this issue ? The `section` string is from `datasets/wikipedia.py:L548` ? You could just add a `try` statement and print the section if the line `section_text.append(section.strip_code().strip())` crashes.
It will help us know if we have to fix it on our side or if it is a `mwparserfromhell` issue. |
https://github.com/huggingface/datasets/issues/321 | ERROR:root:mwparserfromhell | Hi,
Thank you for you answer.
I have try to print the bad section using `try` and `except`, but it is a bit weird as the error seems to appear 3 times for instance, but the two first error does not print anything (as if the function did not go in the `except` part).
For the third one, I got that (I haven't display the entire text) :
> error : ==== Parque nacional Cajas ====
> {{AP|Parque nacional Cajas}}
> [[Archivo:Ecuador cajas national park.jpg|thumb|left|300px|Laguna del Cajas]]
> El parque nacional Cajas está situado en los [[Cordillera de los Andes|Andes]], al sur del [[Ecuador]], en la provincia de [[Provincia de Azuay|Azuay]], a 33
> [[km]] al noroccidente de la ciudad de [[Cuenca (Ecuador)|Cuenca]]. Los accesos más comunes al parque inician todos en Cuenca: Desde allí, la vía Cuenca-Mol
> leturo atraviesa en Control de [[Surocucho]] en poco más de 30 minutos de viaje; más adelante, esta misma carretera pasa a orillas de la laguna La Toreadora donde están el Centro Administrativo y de Información del parque. Siguiendo de largo hacia [[Molleturo]], por esta vía se conoce el sector norte del Cajas y se serpentea entre varias lagunas mayores y menores.
> Para acceder al parque desde la costa, la vía Molleturo-Cuenca es también la mejor opción.
How can I display the link instead of the text ? I suppose it will help you more | Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
| 234 | ERROR:root:mwparserfromhell
Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
Hi,
Thank you for you answer.
I have try to print the bad section using `try` and `except`, but it is a bit weird as the error seems to appear 3 times for instance, but the two first error does not print anything (as if the function did not go in the `except` part).
For the third one, I got that (I haven't display the entire text) :
> error : ==== Parque nacional Cajas ====
> {{AP|Parque nacional Cajas}}
> [[Archivo:Ecuador cajas national park.jpg|thumb|left|300px|Laguna del Cajas]]
> El parque nacional Cajas está situado en los [[Cordillera de los Andes|Andes]], al sur del [[Ecuador]], en la provincia de [[Provincia de Azuay|Azuay]], a 33
> [[km]] al noroccidente de la ciudad de [[Cuenca (Ecuador)|Cuenca]]. Los accesos más comunes al parque inician todos en Cuenca: Desde allí, la vía Cuenca-Mol
> leturo atraviesa en Control de [[Surocucho]] en poco más de 30 minutos de viaje; más adelante, esta misma carretera pasa a orillas de la laguna La Toreadora donde están el Centro Administrativo y de Información del parque. Siguiendo de largo hacia [[Molleturo]], por esta vía se conoce el sector norte del Cajas y se serpentea entre varias lagunas mayores y menores.
> Para acceder al parque desde la costa, la vía Molleturo-Cuenca es también la mejor opción.
How can I display the link instead of the text ? I suppose it will help you more |
https://github.com/huggingface/datasets/issues/321 | ERROR:root:mwparserfromhell | The error appears several times as Apache Beam retries to process examples up to 4 times irc.
I just tried to run this text into `mwparserfromhell` but it worked without the issue.
I used this code (from the `wikipedia.py` script):
```python
import mwparserfromhell as parser
import re
import six
raw_content = r"""==== Parque nacional Cajas ====
{{AP|Parque nacional Cajas}}
[[Archivo:Ecuador cajas national park.jpg|thumb|left|300px|Laguna del Cajas]]
El parque nacional Cajas está situado en los [[Cordillera de los Andes|Andes]], al sur del [[Ecuador]], en la provincia de [[Provincia de Azuay|Azuay]], a 33
[[km]] al noroccidente de la ciudad de [[Cuenca (Ecuador)|Cuenca]]. Los accesos más comunes al parque inician todos en Cuenca: Desde allí, la vía Cuenca-Mol
leturo atraviesa en Control de [[Surocucho]] en poco más de 30 minutos de viaje; más adelante, esta misma carretera pasa a orillas de la laguna La Toreadora donde están el Centro Administrativo y de Información del parque. Siguiendo de largo hacia [[Molleturo]], por esta vía se conoce el sector norte del Cajas y se serpentea entre varias lagunas mayores y menores.
"""
wikicode = parser.parse(raw_content)
# Filters for references, tables, and file/image links.
re_rm_wikilink = re.compile("^(?:File|Image|Media):", flags=re.IGNORECASE | re.UNICODE)
def rm_wikilink(obj):
return bool(re_rm_wikilink.match(six.text_type(obj.title)))
def rm_tag(obj):
return six.text_type(obj.tag) in {"ref", "table"}
def rm_template(obj):
return obj.name.lower() in {"reflist", "notelist", "notelist-ua", "notelist-lr", "notelist-ur", "notelist-lg"}
def try_remove_obj(obj, section):
try:
section.remove(obj)
except ValueError:
# For unknown reasons, objects are sometimes not found.
pass
section_text = []
for section in wikicode.get_sections(flat=True, include_lead=True, include_headings=True):
for obj in section.ifilter_wikilinks(matches=rm_wikilink, recursive=True):
try_remove_obj(obj, section)
for obj in section.ifilter_templates(matches=rm_template, recursive=True):
try_remove_obj(obj, section)
for obj in section.ifilter_tags(matches=rm_tag, recursive=True):
try_remove_obj(obj, section)
section_text.append(section.strip_code().strip())
``` | Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
| 264 | ERROR:root:mwparserfromhell
Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
The error appears several times as Apache Beam retries to process examples up to 4 times irc.
I just tried to run this text into `mwparserfromhell` but it worked without the issue.
I used this code (from the `wikipedia.py` script):
```python
import mwparserfromhell as parser
import re
import six
raw_content = r"""==== Parque nacional Cajas ====
{{AP|Parque nacional Cajas}}
[[Archivo:Ecuador cajas national park.jpg|thumb|left|300px|Laguna del Cajas]]
El parque nacional Cajas está situado en los [[Cordillera de los Andes|Andes]], al sur del [[Ecuador]], en la provincia de [[Provincia de Azuay|Azuay]], a 33
[[km]] al noroccidente de la ciudad de [[Cuenca (Ecuador)|Cuenca]]. Los accesos más comunes al parque inician todos en Cuenca: Desde allí, la vía Cuenca-Mol
leturo atraviesa en Control de [[Surocucho]] en poco más de 30 minutos de viaje; más adelante, esta misma carretera pasa a orillas de la laguna La Toreadora donde están el Centro Administrativo y de Información del parque. Siguiendo de largo hacia [[Molleturo]], por esta vía se conoce el sector norte del Cajas y se serpentea entre varias lagunas mayores y menores.
"""
wikicode = parser.parse(raw_content)
# Filters for references, tables, and file/image links.
re_rm_wikilink = re.compile("^(?:File|Image|Media):", flags=re.IGNORECASE | re.UNICODE)
def rm_wikilink(obj):
return bool(re_rm_wikilink.match(six.text_type(obj.title)))
def rm_tag(obj):
return six.text_type(obj.tag) in {"ref", "table"}
def rm_template(obj):
return obj.name.lower() in {"reflist", "notelist", "notelist-ua", "notelist-lr", "notelist-ur", "notelist-lg"}
def try_remove_obj(obj, section):
try:
section.remove(obj)
except ValueError:
# For unknown reasons, objects are sometimes not found.
pass
section_text = []
for section in wikicode.get_sections(flat=True, include_lead=True, include_headings=True):
for obj in section.ifilter_wikilinks(matches=rm_wikilink, recursive=True):
try_remove_obj(obj, section)
for obj in section.ifilter_templates(matches=rm_template, recursive=True):
try_remove_obj(obj, section)
for obj in section.ifilter_tags(matches=rm_tag, recursive=True):
try_remove_obj(obj, section)
section_text.append(section.strip_code().strip())
``` |
https://github.com/huggingface/datasets/issues/321 | ERROR:root:mwparserfromhell | Not sure why we're having this issue. Maybe could you get also the file that's causing that ? | Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
| 18 | ERROR:root:mwparserfromhell
Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
Not sure why we're having this issue. Maybe could you get also the file that's causing that ? |
https://github.com/huggingface/datasets/issues/321 | ERROR:root:mwparserfromhell | thanks for your answer.
How can I know which file is causing the issue ?
I am trying to load the spanish wikipedia data. | Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
| 24 | ERROR:root:mwparserfromhell
Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
thanks for your answer.
How can I know which file is causing the issue ?
I am trying to load the spanish wikipedia data. |
https://github.com/huggingface/datasets/issues/321 | ERROR:root:mwparserfromhell | Because of the way Apache Beam works we indeed don't have access to the file name at this point in the code.
We'll have to use some tricks I think :p
You can append `filepath` to `title` in `wikipedia.py:L512` for example. [[EDIT: it's L494 my bad]]
Then just do `try:...except:` on the call of `_parse_and_clean_wikicode` L500 I guess.
Thanks for diving into this ! I tried it myself but I run out of memory on my laptop
As soon as we have the name of the file it should be easier to find what's wrong. | Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
| 95 | ERROR:root:mwparserfromhell
Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
Because of the way Apache Beam works we indeed don't have access to the file name at this point in the code.
We'll have to use some tricks I think :p
You can append `filepath` to `title` in `wikipedia.py:L512` for example. [[EDIT: it's L494 my bad]]
Then just do `try:...except:` on the call of `_parse_and_clean_wikicode` L500 I guess.
Thanks for diving into this ! I tried it myself but I run out of memory on my laptop
As soon as we have the name of the file it should be easier to find what's wrong. |
https://github.com/huggingface/datasets/issues/321 | ERROR:root:mwparserfromhell | Thanks for your help.
I tried to print the "title" of the document inside the` except (mwparserfromhell.parser.ParserError) as e`,the title displayed was : "Campeonato Mundial de futsal de la AMF 2015". (Wikipedia ES) Is it what you were looking for ? | Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
| 41 | ERROR:root:mwparserfromhell
Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
Thanks for your help.
I tried to print the "title" of the document inside the` except (mwparserfromhell.parser.ParserError) as e`,the title displayed was : "Campeonato Mundial de futsal de la AMF 2015". (Wikipedia ES) Is it what you were looking for ? |
https://github.com/huggingface/datasets/issues/321 | ERROR:root:mwparserfromhell | Thanks a lot @Shiro-LK !
I was able to reproduce the issue. It comes from [this table on wikipedia](https://es.wikipedia.org/wiki/Campeonato_Mundial_de_futsal_de_la_AMF_2015#Clasificados) that can't be parsed.
The file in which the problem occurs comes from the wikipedia dumps, and it can be downloaded [here](https://dumps.wikimedia.org/eswiki/20200501/eswiki-20200501-pages-articles-multistream6.xml-p6424816p7924815.bz2)
Parsing the file this way raises the parsing issue:
```python
import mwparserfromhell as parser
from tqdm.auto import tqdm
import bz2
import six
import logging
import codecs
import xml.etree.cElementTree as etree
filepath = "path/to/eswiki-20200501-pages-articles-multistream6.xml-p6424816p7924815.bz2"
def _extract_content(filepath):
"""Extracts article content from a single WikiMedia XML file."""
logging.info("generating examples from = %s", filepath)
with open(filepath, "rb") as f:
f = bz2.BZ2File(filename=f)
if six.PY3:
# Workaround due to:
# https://github.com/tensorflow/tensorflow/issues/33563
utf_f = codecs.getreader("utf-8")(f)
else:
utf_f = f
# To clear root, to free-up more memory than just `elem.clear()`.
context = etree.iterparse(utf_f, events=("end",))
context = iter(context)
unused_event, root = next(context)
for unused_event, elem in tqdm(context, total=949087):
if not elem.tag.endswith("page"):
continue
namespace = elem.tag[:-4]
title = elem.find("./{0}title".format(namespace)).text
ns = elem.find("./{0}ns".format(namespace)).text
id_ = elem.find("./{0}id".format(namespace)).text
# Filter pages that are not in the "main" namespace.
if ns != "0":
root.clear()
continue
raw_content = elem.find("./{0}revision/{0}text".format(namespace)).text
root.clear()
if "Campeonato Mundial de futsal de la AMF 2015" in title:
yield (id_, title, raw_content)
for id_, title, raw_content in _extract_content(filepath):
wikicode = parser.parse(raw_content)
```
The copied the raw content that can't be parsed [here](https://pastebin.com/raw/ZbmevLyH).
The minimal code to reproduce is:
```python
import mwparserfromhell as parser
import requests
raw_content = requests.get("https://pastebin.com/raw/ZbmevLyH").content.decode("utf-8")
wikicode = parser.parse(raw_content)
```
I will create an issue on mwparserfromhell's repo to see if we can fix that
| Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
| 248 | ERROR:root:mwparserfromhell
Hi,
I am trying to download some wikipedia data but I got this error for spanish "es" (but there are maybe some others languages which have the same error I haven't tried all of them ).
`ERROR:root:mwparserfromhell ParseError: This is a bug and should be reported. Info: C tokenizer exited with non-empty token stack.`
The code I have use was :
`dataset = load_dataset('wikipedia', '20200501.es', beam_runner='DirectRunner')`
Thanks a lot @Shiro-LK !
I was able to reproduce the issue. It comes from [this table on wikipedia](https://es.wikipedia.org/wiki/Campeonato_Mundial_de_futsal_de_la_AMF_2015#Clasificados) that can't be parsed.
The file in which the problem occurs comes from the wikipedia dumps, and it can be downloaded [here](https://dumps.wikimedia.org/eswiki/20200501/eswiki-20200501-pages-articles-multistream6.xml-p6424816p7924815.bz2)
Parsing the file this way raises the parsing issue:
```python
import mwparserfromhell as parser
from tqdm.auto import tqdm
import bz2
import six
import logging
import codecs
import xml.etree.cElementTree as etree
filepath = "path/to/eswiki-20200501-pages-articles-multistream6.xml-p6424816p7924815.bz2"
def _extract_content(filepath):
"""Extracts article content from a single WikiMedia XML file."""
logging.info("generating examples from = %s", filepath)
with open(filepath, "rb") as f:
f = bz2.BZ2File(filename=f)
if six.PY3:
# Workaround due to:
# https://github.com/tensorflow/tensorflow/issues/33563
utf_f = codecs.getreader("utf-8")(f)
else:
utf_f = f
# To clear root, to free-up more memory than just `elem.clear()`.
context = etree.iterparse(utf_f, events=("end",))
context = iter(context)
unused_event, root = next(context)
for unused_event, elem in tqdm(context, total=949087):
if not elem.tag.endswith("page"):
continue
namespace = elem.tag[:-4]
title = elem.find("./{0}title".format(namespace)).text
ns = elem.find("./{0}ns".format(namespace)).text
id_ = elem.find("./{0}id".format(namespace)).text
# Filter pages that are not in the "main" namespace.
if ns != "0":
root.clear()
continue
raw_content = elem.find("./{0}revision/{0}text".format(namespace)).text
root.clear()
if "Campeonato Mundial de futsal de la AMF 2015" in title:
yield (id_, title, raw_content)
for id_, title, raw_content in _extract_content(filepath):
wikicode = parser.parse(raw_content)
```
The copied the raw content that can't be parsed [here](https://pastebin.com/raw/ZbmevLyH).
The minimal code to reproduce is:
```python
import mwparserfromhell as parser
import requests
raw_content = requests.get("https://pastebin.com/raw/ZbmevLyH").content.decode("utf-8")
wikicode = parser.parse(raw_content)
```
I will create an issue on mwparserfromhell's repo to see if we can fix that
|
https://github.com/huggingface/datasets/issues/320 | Blog Authorship Corpus, Non Matching Splits Sizes Error, nlp viewer | This dataset seems to have a decoding problem that results in inconsistencies in the number of generated examples.
See #215.
That's why we end up with a `NonMatchingSplitsSizesError `. | Selecting `blog_authorship_corpus` in the nlp viewer throws the following error:
```
NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='train', num_bytes=610252351, num_examples=532812, dataset_name='blog_authorship_corpus'), 'recorded': SplitInfo(name='train', num_bytes=614706451, num_examples=535568, dataset_name='blog_authorship_corpus')}, {'expected': SplitInfo(name='validation', num_bytes=37500394, num_examples=31277, dataset_name='blog_authorship_corpus'), 'recorded': SplitInfo(name='validation', num_bytes=32553710, num_examples=28521, dataset_name='blog_authorship_corpus')}]
Traceback:
File "/home/sasha/streamlit/lib/streamlit/ScriptRunner.py", line 322, in _run_script
exec(code, module.__dict__)
File "/home/sasha/nlp-viewer/run.py", line 172, in <module>
dts, fail = get(str(option.id), str(conf_option.name) if conf_option else None)
File "/home/sasha/streamlit/lib/streamlit/caching.py", line 591, in wrapped_func
return get_or_create_cached_value()
File "/home/sasha/streamlit/lib/streamlit/caching.py", line 575, in get_or_create_cached_value
return_value = func(*args, **kwargs)
File "/home/sasha/nlp-viewer/run.py", line 132, in get
builder_instance.download_and_prepare()
File "/home/sasha/.local/share/virtualenvs/lib-ogGKnCK_/lib/python3.7/site-packages/nlp/builder.py", line 432, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/home/sasha/.local/share/virtualenvs/lib-ogGKnCK_/lib/python3.7/site-packages/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/home/sasha/.local/share/virtualenvs/lib-ogGKnCK_/lib/python3.7/site-packages/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
```
@srush @lhoestq | 29 | Blog Authorship Corpus, Non Matching Splits Sizes Error, nlp viewer
Selecting `blog_authorship_corpus` in the nlp viewer throws the following error:
```
NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='train', num_bytes=610252351, num_examples=532812, dataset_name='blog_authorship_corpus'), 'recorded': SplitInfo(name='train', num_bytes=614706451, num_examples=535568, dataset_name='blog_authorship_corpus')}, {'expected': SplitInfo(name='validation', num_bytes=37500394, num_examples=31277, dataset_name='blog_authorship_corpus'), 'recorded': SplitInfo(name='validation', num_bytes=32553710, num_examples=28521, dataset_name='blog_authorship_corpus')}]
Traceback:
File "/home/sasha/streamlit/lib/streamlit/ScriptRunner.py", line 322, in _run_script
exec(code, module.__dict__)
File "/home/sasha/nlp-viewer/run.py", line 172, in <module>
dts, fail = get(str(option.id), str(conf_option.name) if conf_option else None)
File "/home/sasha/streamlit/lib/streamlit/caching.py", line 591, in wrapped_func
return get_or_create_cached_value()
File "/home/sasha/streamlit/lib/streamlit/caching.py", line 575, in get_or_create_cached_value
return_value = func(*args, **kwargs)
File "/home/sasha/nlp-viewer/run.py", line 132, in get
builder_instance.download_and_prepare()
File "/home/sasha/.local/share/virtualenvs/lib-ogGKnCK_/lib/python3.7/site-packages/nlp/builder.py", line 432, in download_and_prepare
dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
File "/home/sasha/.local/share/virtualenvs/lib-ogGKnCK_/lib/python3.7/site-packages/nlp/builder.py", line 488, in _download_and_prepare
verify_splits(self.info.splits, split_dict)
File "/home/sasha/.local/share/virtualenvs/lib-ogGKnCK_/lib/python3.7/site-packages/nlp/utils/info_utils.py", line 70, in verify_splits
raise NonMatchingSplitsSizesError(str(bad_splits))
```
@srush @lhoestq
This dataset seems to have a decoding problem that results in inconsistencies in the number of generated examples.
See #215.
That's why we end up with a `NonMatchingSplitsSizesError `. |
https://github.com/huggingface/datasets/issues/319 | Nested sequences with dicts | Oh yes, this is a backward compatibility feature with tensorflow_dataset in which a `Sequence` or `dict` is converted in a `dict` of `lists`, unfortunately it is not very intuitive, see here: https://github.com/huggingface/nlp/blob/master/src/nlp/features.py#L409
To avoid this behavior, you can just define the list in the feature with a simple list or a tuple (which is also simpler to write).
In your case, the features could be as follow:
``` python
...
features=nlp.Features({
"title": nlp.Value("string"),
"vertexSet": [[{
"name": nlp.Value("string"),
"sent_id": nlp.Value("int32"),
"pos": nlp.features.Sequence(nlp.Value("int32")),
"type": nlp.Value("string"),
}]],
...
}),
...
``` | Am pretty much finished [adding a dataset](https://github.com/ghomasHudson/nlp/blob/DocRED/datasets/docred/docred.py) for [DocRED](https://github.com/thunlp/DocRED), but am getting an error when trying to add a nested `nlp.features.sequence(nlp.features.sequence({key:value,...}))`.
The original data is in this format:
```python
{
'title': "Title of wiki page",
'vertexSet': [
[
{ 'name': "mention_name",
'sent_id': "mention in which sentence",
'pos': ["postion of mention in a sentence"],
'type': "NER_type"},
{another mention}
],
[another entity]
]
...
}
```
So to represent this I've attempted to write:
```
...
features=nlp.Features({
"title": nlp.Value("string"),
"vertexSet": nlp.features.Sequence(nlp.features.Sequence({
"name": nlp.Value("string"),
"sent_id": nlp.Value("int32"),
"pos": nlp.features.Sequence(nlp.Value("int32")),
"type": nlp.Value("string"),
})),
...
}),
...
```
This is giving me the error:
```
pyarrow.lib.ArrowTypeError: Could not convert [{'pos': [[0,2], [2,4], [3,5]], "type": ["ORG", "ORG", "ORG"], "name": ["Lark Force", "Lark Force", "Lark Force", "sent_id": [0, 3, 4]}..... with type list: was not a dict, tuple, or recognized null value for conversion to struct type
```
Do we expect the pyarrow stuff to break when doing this deeper nesting? I've checked that it still works when you do `nlp.features.Sequence(nlp.features.Sequence(nlp.Value("string"))` or `nlp.features.Sequence({key:value,...})` just not nested sequences with a dict.
If it's not possible, I can always convert it to a shallower structure. I'd rather not change the DocRED authors' structure if I don't have to though. | 88 | Nested sequences with dicts
Am pretty much finished [adding a dataset](https://github.com/ghomasHudson/nlp/blob/DocRED/datasets/docred/docred.py) for [DocRED](https://github.com/thunlp/DocRED), but am getting an error when trying to add a nested `nlp.features.sequence(nlp.features.sequence({key:value,...}))`.
The original data is in this format:
```python
{
'title': "Title of wiki page",
'vertexSet': [
[
{ 'name': "mention_name",
'sent_id': "mention in which sentence",
'pos': ["postion of mention in a sentence"],
'type': "NER_type"},
{another mention}
],
[another entity]
]
...
}
```
So to represent this I've attempted to write:
```
...
features=nlp.Features({
"title": nlp.Value("string"),
"vertexSet": nlp.features.Sequence(nlp.features.Sequence({
"name": nlp.Value("string"),
"sent_id": nlp.Value("int32"),
"pos": nlp.features.Sequence(nlp.Value("int32")),
"type": nlp.Value("string"),
})),
...
}),
...
```
This is giving me the error:
```
pyarrow.lib.ArrowTypeError: Could not convert [{'pos': [[0,2], [2,4], [3,5]], "type": ["ORG", "ORG", "ORG"], "name": ["Lark Force", "Lark Force", "Lark Force", "sent_id": [0, 3, 4]}..... with type list: was not a dict, tuple, or recognized null value for conversion to struct type
```
Do we expect the pyarrow stuff to break when doing this deeper nesting? I've checked that it still works when you do `nlp.features.Sequence(nlp.features.Sequence(nlp.Value("string"))` or `nlp.features.Sequence({key:value,...})` just not nested sequences with a dict.
If it's not possible, I can always convert it to a shallower structure. I'd rather not change the DocRED authors' structure if I don't have to though.
Oh yes, this is a backward compatibility feature with tensorflow_dataset in which a `Sequence` or `dict` is converted in a `dict` of `lists`, unfortunately it is not very intuitive, see here: https://github.com/huggingface/nlp/blob/master/src/nlp/features.py#L409
To avoid this behavior, you can just define the list in the feature with a simple list or a tuple (which is also simpler to write).
In your case, the features could be as follow:
``` python
...
features=nlp.Features({
"title": nlp.Value("string"),
"vertexSet": [[{
"name": nlp.Value("string"),
"sent_id": nlp.Value("int32"),
"pos": nlp.features.Sequence(nlp.Value("int32")),
"type": nlp.Value("string"),
}]],
...
}),
...
``` |
https://github.com/huggingface/datasets/issues/317 | Adding a dataset with multiple subtasks | For one dataset you can have different configurations that each have their own `nlp.Features`.
We imagine having one configuration per subtask for example.
They are loaded with `nlp.load_dataset("my_dataset", "my_config")`.
For example the `glue` dataset has many configurations. It is a bit different from your case though because each configuration is a dataset by itself (sst2, mnli).
Another example is `wikipedia` that has one configuration per language. | I intent to add the datasets of the MT Quality Estimation shared tasks to `nlp`. However, they have different subtasks -- such as word-level, sentence-level and document-level quality estimation, each of which having different language pairs, and some of the data reused in different subtasks.
For example, in [QE 2019,](http://www.statmt.org/wmt19/qe-task.html) we had the same English-Russian and English-German data for word-level and sentence-level QE.
I suppose these datasets could have both their word and sentence-level labels inside `nlp.Features`; but what about other subtasks? Should they be considered a different dataset altogether?
I read the discussion on #217 but the case of QE seems a lot simpler. | 66 | Adding a dataset with multiple subtasks
I intent to add the datasets of the MT Quality Estimation shared tasks to `nlp`. However, they have different subtasks -- such as word-level, sentence-level and document-level quality estimation, each of which having different language pairs, and some of the data reused in different subtasks.
For example, in [QE 2019,](http://www.statmt.org/wmt19/qe-task.html) we had the same English-Russian and English-German data for word-level and sentence-level QE.
I suppose these datasets could have both their word and sentence-level labels inside `nlp.Features`; but what about other subtasks? Should they be considered a different dataset altogether?
I read the discussion on #217 but the case of QE seems a lot simpler.
For one dataset you can have different configurations that each have their own `nlp.Features`.
We imagine having one configuration per subtask for example.
They are loaded with `nlp.load_dataset("my_dataset", "my_config")`.
For example the `glue` dataset has many configurations. It is a bit different from your case though because each configuration is a dataset by itself (sst2, mnli).
Another example is `wikipedia` that has one configuration per language. |
https://github.com/huggingface/datasets/issues/315 | [Question] Best way to batch a large dataset? | Update: I think I've found a solution.
```python
output_types = {"input_ids": tf.int64, "token_type_ids": tf.int64, "attention_mask": tf.int64}
def train_dataset_gen():
for i in range(len(train_dataset)):
yield train_dataset[i]
tf_dataset = tf.data.Dataset.from_generator(train_dataset_gen, output_types=output_types)
```
loads WikiText-2 in 20 ms, and WikiText-103 in 20 ms. It appears to be lazily loading via indexing train_dataset. | I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets? | 48 | [Question] Best way to batch a large dataset?
I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets?
Update: I think I've found a solution.
```python
output_types = {"input_ids": tf.int64, "token_type_ids": tf.int64, "attention_mask": tf.int64}
def train_dataset_gen():
for i in range(len(train_dataset)):
yield train_dataset[i]
tf_dataset = tf.data.Dataset.from_generator(train_dataset_gen, output_types=output_types)
```
loads WikiText-2 in 20 ms, and WikiText-103 in 20 ms. It appears to be lazily loading via indexing train_dataset. |
https://github.com/huggingface/datasets/issues/315 | [Question] Best way to batch a large dataset? | Yes this is the current best solution. We should probably show it in the tutorial notebook.
Note that this solution unfortunately doesn't allow to train on TPUs (yet). See #193 | I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets? | 30 | [Question] Best way to batch a large dataset?
I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets?
Yes this is the current best solution. We should probably show it in the tutorial notebook.
Note that this solution unfortunately doesn't allow to train on TPUs (yet). See #193 |
https://github.com/huggingface/datasets/issues/315 | [Question] Best way to batch a large dataset? | This approach still seems quite slow. When using TFRecords with a similar training loop, I get ~3.0-3.5 it/s on multi-node, multi-GPU training. I notice a pretty severe performance regression when scaling, with observed performance numbers. Since the allreduce step takes less than 100ms/it and I've achieved 80% scaling efficiency up to 64 GPUs, it must be the data pipeline.
| Nodes | GPUs | Iterations/Second |
| --- | --- | --- |
| 1 | 2 | 2.01 |
| 1 | 8 | 0.81 |
| 2 | 16 | 0.37 |
Here are performance metrics over 10k steps. The iteration speed appears to follow some sort of caching pattern. I would love to use `nlp` in my project, but a slowdown from 3.0 it/s to 0.3 it/s is too great to stomach.
<img width="1361" alt="Screen Shot 2020-07-02 at 8 29 22 AM" src="https://user-images.githubusercontent.com/4564897/86378156-2f8d3900-bc3e-11ea-918b-c395c3df5377.png">
| I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets? | 146 | [Question] Best way to batch a large dataset?
I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets?
This approach still seems quite slow. When using TFRecords with a similar training loop, I get ~3.0-3.5 it/s on multi-node, multi-GPU training. I notice a pretty severe performance regression when scaling, with observed performance numbers. Since the allreduce step takes less than 100ms/it and I've achieved 80% scaling efficiency up to 64 GPUs, it must be the data pipeline.
| Nodes | GPUs | Iterations/Second |
| --- | --- | --- |
| 1 | 2 | 2.01 |
| 1 | 8 | 0.81 |
| 2 | 16 | 0.37 |
Here are performance metrics over 10k steps. The iteration speed appears to follow some sort of caching pattern. I would love to use `nlp` in my project, but a slowdown from 3.0 it/s to 0.3 it/s is too great to stomach.
<img width="1361" alt="Screen Shot 2020-07-02 at 8 29 22 AM" src="https://user-images.githubusercontent.com/4564897/86378156-2f8d3900-bc3e-11ea-918b-c395c3df5377.png">
|
https://github.com/huggingface/datasets/issues/315 | [Question] Best way to batch a large dataset? | An interesting alternative to investigate here would be to use the tf.io library which has some support for Arrow to TF conversion: https://www.tensorflow.org/io/api_docs/python/tfio/arrow/ArrowDataset
There are quite a few types supported, including lists so if the unsupported columns are dropped then we could maybe have a zero-copy mapping from Arrow to TensorFlow, including tokenized inputs and 1D tensors like the ones we mostly use in NLP: https://github.com/tensorflow/io/blob/322b3170c43ecac5c6af9e39dbd18fd747913e5a/tensorflow_io/arrow/python/ops/arrow_dataset_ops.py#L44-L72
Here is an introduction on Arrow to TF using tf.io: https://medium.com/tensorflow/tensorflow-with-apache-arrow-datasets-cdbcfe80a59f | I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets? | 77 | [Question] Best way to batch a large dataset?
I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets?
An interesting alternative to investigate here would be to use the tf.io library which has some support for Arrow to TF conversion: https://www.tensorflow.org/io/api_docs/python/tfio/arrow/ArrowDataset
There are quite a few types supported, including lists so if the unsupported columns are dropped then we could maybe have a zero-copy mapping from Arrow to TensorFlow, including tokenized inputs and 1D tensors like the ones we mostly use in NLP: https://github.com/tensorflow/io/blob/322b3170c43ecac5c6af9e39dbd18fd747913e5a/tensorflow_io/arrow/python/ops/arrow_dataset_ops.py#L44-L72
Here is an introduction on Arrow to TF using tf.io: https://medium.com/tensorflow/tensorflow-with-apache-arrow-datasets-cdbcfe80a59f |
https://github.com/huggingface/datasets/issues/315 | [Question] Best way to batch a large dataset? | Interesting. There's no support for strings, but it does enable int and floats so that would work for tokenized inputs.
ArrowStreamDataset requires loading from a "record batch iterator", which can be instantiated from in-memory arrays as described here: https://arrow.apache.org/docs/python/ipc.html.
But the nlp.Dataset stores its data as a `pyarrow.lib.Table`, and the underlying features are `pyarrow.lib.ChunkedArray`. I can't find any documentation about lazily creating a record batch iterator from a ChunkedArray or a Table. Have you had any success?
I can't find [any uses](https://grep.app/search?q=ArrowDataset&filter[lang][0]=Python) of tfio.arrow.ArrowDataset on GitHub. | I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets? | 86 | [Question] Best way to batch a large dataset?
I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets?
Interesting. There's no support for strings, but it does enable int and floats so that would work for tokenized inputs.
ArrowStreamDataset requires loading from a "record batch iterator", which can be instantiated from in-memory arrays as described here: https://arrow.apache.org/docs/python/ipc.html.
But the nlp.Dataset stores its data as a `pyarrow.lib.Table`, and the underlying features are `pyarrow.lib.ChunkedArray`. I can't find any documentation about lazily creating a record batch iterator from a ChunkedArray or a Table. Have you had any success?
I can't find [any uses](https://grep.app/search?q=ArrowDataset&filter[lang][0]=Python) of tfio.arrow.ArrowDataset on GitHub. |
https://github.com/huggingface/datasets/issues/315 | [Question] Best way to batch a large dataset? | Also note that since #322 it is now possible to do
```python
ids = [1, 10, 42, 100]
batch = dataset[ids]
```
From my experience it is quite fast but it can take lots of memory for large batches (haven't played that much with it).
Let me know if you think there could be a better way to implement it. (current code is [here](https://github.com/huggingface/nlp/blob/78628649962671b4aaa31a6b24e7275533416845/src/nlp/arrow_dataset.py#L463)) | I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets? | 64 | [Question] Best way to batch a large dataset?
I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets?
Also note that since #322 it is now possible to do
```python
ids = [1, 10, 42, 100]
batch = dataset[ids]
```
From my experience it is quite fast but it can take lots of memory for large batches (haven't played that much with it).
Let me know if you think there could be a better way to implement it. (current code is [here](https://github.com/huggingface/nlp/blob/78628649962671b4aaa31a6b24e7275533416845/src/nlp/arrow_dataset.py#L463)) |
https://github.com/huggingface/datasets/issues/315 | [Question] Best way to batch a large dataset? | Thanks @lhoestq! That format is much better to work with.
I put together a benchmarking script. This doesn't measure the CPU-to-GPU efficiency, nor how it scales with multi-GPU multi-node training where many processes are making the same demands on the same dataset. But it does show some interesting results:
```python
import nlp
import numpy as np
import tensorflow as tf
import time
dset = nlp.load_dataset("wikitext", "wikitext-2-raw-v1", split="train")
dset = dset.filter(lambda ex: len(ex["text"]) > 0)
bsz = 1024
n_batches = 100
def single_item_gen():
for i in range(len(dset)):
yield dset[i]
def sequential_batch_gen():
for i in range(0, len(dset), bsz):
yield dset[i:i+bsz]
def random_batch_gen():
for i in range(len(dset)):
indices = list(np.random.randint(len(dset), size=(bsz,)))
yield dset[indices]
output_types = {"text": tf.string}
single_item = tf.data.Dataset.from_generator(single_item_gen, output_types=output_types).batch(bsz)
interleaved = tf.data.Dataset.range(10).interleave(
lambda idx: tf.data.Dataset.from_generator(single_item_gen, output_types=output_types),
cycle_length=10,
)
sequential_batch = tf.data.Dataset.from_generator(sequential_batch_gen, output_types=output_types)
random_batch = tf.data.Dataset.from_generator(random_batch_gen, output_types=output_types)
def iterate(tf_dset):
start = time.perf_counter()
for i, batch in enumerate(tf_dset.take(n_batches)):
pass
elapsed = time.perf_counter() - start
print(f"{tf_dset} took {elapsed:.3f} secs")
iterate(single_item)
iterate(interleaved)
iterate(sequential_batch)
iterate(random_batch)
```
Results:
```
<BatchDataset shapes: {text: <unknown>}, types: {text: tf.string}> took 23.005 secs
<InterleaveDataset shapes: {text: <unknown>}, types: {text: tf.string}> took 0.135 secs
<FlatMapDataset shapes: {text: <unknown>}, types: {text: tf.string}> took 0.074 secs
<FlatMapDataset shapes: {text: <unknown>}, types: {text: tf.string}> took 0.550 secs
```
- Batching a generator which fetches a single item is terrible.
- Interleaving performs well on a single process, but doesn't scale well to multi-GPU training. I believe the bottleneck here is in Arrow dataset locking or something similar. The numbers from the table above are with interleaving.
- The sequential access dominates the random access (7x faster). Is there any way to bring random access times closer to sequential access? Maybe re-indexing the dataset after shuffling each pass over the data. | I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets? | 285 | [Question] Best way to batch a large dataset?
I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets?
Thanks @lhoestq! That format is much better to work with.
I put together a benchmarking script. This doesn't measure the CPU-to-GPU efficiency, nor how it scales with multi-GPU multi-node training where many processes are making the same demands on the same dataset. But it does show some interesting results:
```python
import nlp
import numpy as np
import tensorflow as tf
import time
dset = nlp.load_dataset("wikitext", "wikitext-2-raw-v1", split="train")
dset = dset.filter(lambda ex: len(ex["text"]) > 0)
bsz = 1024
n_batches = 100
def single_item_gen():
for i in range(len(dset)):
yield dset[i]
def sequential_batch_gen():
for i in range(0, len(dset), bsz):
yield dset[i:i+bsz]
def random_batch_gen():
for i in range(len(dset)):
indices = list(np.random.randint(len(dset), size=(bsz,)))
yield dset[indices]
output_types = {"text": tf.string}
single_item = tf.data.Dataset.from_generator(single_item_gen, output_types=output_types).batch(bsz)
interleaved = tf.data.Dataset.range(10).interleave(
lambda idx: tf.data.Dataset.from_generator(single_item_gen, output_types=output_types),
cycle_length=10,
)
sequential_batch = tf.data.Dataset.from_generator(sequential_batch_gen, output_types=output_types)
random_batch = tf.data.Dataset.from_generator(random_batch_gen, output_types=output_types)
def iterate(tf_dset):
start = time.perf_counter()
for i, batch in enumerate(tf_dset.take(n_batches)):
pass
elapsed = time.perf_counter() - start
print(f"{tf_dset} took {elapsed:.3f} secs")
iterate(single_item)
iterate(interleaved)
iterate(sequential_batch)
iterate(random_batch)
```
Results:
```
<BatchDataset shapes: {text: <unknown>}, types: {text: tf.string}> took 23.005 secs
<InterleaveDataset shapes: {text: <unknown>}, types: {text: tf.string}> took 0.135 secs
<FlatMapDataset shapes: {text: <unknown>}, types: {text: tf.string}> took 0.074 secs
<FlatMapDataset shapes: {text: <unknown>}, types: {text: tf.string}> took 0.550 secs
```
- Batching a generator which fetches a single item is terrible.
- Interleaving performs well on a single process, but doesn't scale well to multi-GPU training. I believe the bottleneck here is in Arrow dataset locking or something similar. The numbers from the table above are with interleaving.
- The sequential access dominates the random access (7x faster). Is there any way to bring random access times closer to sequential access? Maybe re-indexing the dataset after shuffling each pass over the data. |
https://github.com/huggingface/datasets/issues/315 | [Question] Best way to batch a large dataset? | Hey @jarednielsen
Thanks for this very interesting analysis!! IMHO to read text data one should use `tf.data.TextLineDataset`. It would be interesting to compare what you have done with simply load with a `TextLineDataset` and see if there is a difference.
A good example can be found here https://www.tensorflow.org/tutorials/load_data/text | I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets? | 48 | [Question] Best way to batch a large dataset?
I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets?
Hey @jarednielsen
Thanks for this very interesting analysis!! IMHO to read text data one should use `tf.data.TextLineDataset`. It would be interesting to compare what you have done with simply load with a `TextLineDataset` and see if there is a difference.
A good example can be found here https://www.tensorflow.org/tutorials/load_data/text |
https://github.com/huggingface/datasets/issues/315 | [Question] Best way to batch a large dataset? | Thanks! I'm not actually loading in raw text data, that was just the synthetic data I created for this benchmark. A more realistic use case would be a dataset of tokenized examples, which would be a dict of lists of integers. TensorFlow's TextLineDataset greedily loads the dataset into the graph itself, which can lead to out-of-memory errors - one of the main reason I'm so drawn to the `nlp` library is its zero-copy no-RAM approach to dataset loading and mapping.
It's quite helpful for running a preprocessing pipeline - a sample ELECTRA pipeline I've built is here: https://github.com/jarednielsen/deep-learning-models/blob/nlp/models/nlp/common/preprocess.py. | I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets? | 98 | [Question] Best way to batch a large dataset?
I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets?
Thanks! I'm not actually loading in raw text data, that was just the synthetic data I created for this benchmark. A more realistic use case would be a dataset of tokenized examples, which would be a dict of lists of integers. TensorFlow's TextLineDataset greedily loads the dataset into the graph itself, which can lead to out-of-memory errors - one of the main reason I'm so drawn to the `nlp` library is its zero-copy no-RAM approach to dataset loading and mapping.
It's quite helpful for running a preprocessing pipeline - a sample ELECTRA pipeline I've built is here: https://github.com/jarednielsen/deep-learning-models/blob/nlp/models/nlp/common/preprocess.py. |
https://github.com/huggingface/datasets/issues/315 | [Question] Best way to batch a large dataset? | Sorry, I think I badly expressed myself, my bad. What I suggested is to compare with the usual loading textual data in pure TF with `TextLineDataset` with `nlp`. I know it is not recommended with very large datasets to use it, but I was curious to see how it behaves compared to a processing with `nlp` on smaller datasets.
BTW your script looks very interesting, thanks for sharing!! | I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets? | 68 | [Question] Best way to batch a large dataset?
I'm training on large datasets such as Wikipedia and BookCorpus. Following the instructions in [the tutorial notebook](https://colab.research.google.com/github/huggingface/nlp/blob/master/notebooks/Overview.ipynb), I see the following recommended for TensorFlow:
```python
train_tf_dataset = train_tf_dataset.filter(remove_none_values, load_from_cache_file=False)
columns = ['input_ids', 'token_type_ids', 'attention_mask', 'start_positions', 'end_positions']
train_tf_dataset.set_format(type='tensorflow', columns=columns)
features = {x: train_tf_dataset[x].to_tensor(default_value=0, shape=[None, tokenizer.max_len]) for x in columns[:3]}
labels = {"output_1": train_tf_dataset["start_positions"].to_tensor(default_value=0, shape=[None, 1])}
labels["output_2"] = train_tf_dataset["end_positions"].to_tensor(default_value=0, shape=[None, 1])
### Question about this last line ###
tfdataset = tf.data.Dataset.from_tensor_slices((features, labels)).batch(8)
```
This code works for something like WikiText-2. However, scaling up to WikiText-103, the last line takes 5-10 minutes to run. I assume it is because tf.data.Dataset.from_tensor_slices() is pulling everything into memory, not lazily loading. This approach won't scale up to datasets 25x larger such as Wikipedia.
So I tried manual batching using `dataset.select()`:
```python
idxs = np.random.randint(len(dataset), size=bsz)
batch = dataset.select(idxs).map(lambda example: {"input_ids": tokenizer(example["text"])})
tf_batch = tf.constant(batch["ids"], dtype=tf.int64)
```
This appears to create a new Apache Arrow dataset with every batch I grab, and then tries to cache it. The runtime of `dataset.select([0, 1])` appears to be much worse than `dataset[:2]`. So using `select()` doesn't seem to be performant enough for a training loop.
Is there a performant scalable way to lazily load batches of nlp Datasets?
Sorry, I think I badly expressed myself, my bad. What I suggested is to compare with the usual loading textual data in pure TF with `TextLineDataset` with `nlp`. I know it is not recommended with very large datasets to use it, but I was curious to see how it behaves compared to a processing with `nlp` on smaller datasets.
BTW your script looks very interesting, thanks for sharing!! |
https://github.com/huggingface/datasets/issues/312 | [Feature request] Add `shard()` method to dataset | Hi Jared,
Interesting, thanks for raising this question. You can also do that after loading with `dataset.select()` or `dataset.filter()` which let you keep only a specific subset of rows in a dataset.
What is your use-case for sharding? | Currently, to shard a dataset into 10 pieces on different ranks, you can run
```python
rank = 3 # for example
size = 10
dataset = nlp.load_dataset('wikitext', 'wikitext-2-raw-v1', split=f"train[{rank*10}%:{(rank+1)*10}%]")
```
However, this breaks down if you have a number of ranks that doesn't divide cleanly into 100, such as 64 ranks. Is there interest in adding a method shard() that looks like this?
```python
rank = 3
size = 64
dataset = nlp.load_dataset("wikitext", "wikitext-2-raw-v1", split="train").shard(rank=rank, size=size)
```
TensorFlow has a similar API: https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shard. I'd be happy to contribute this code. | 38 | [Feature request] Add `shard()` method to dataset
Currently, to shard a dataset into 10 pieces on different ranks, you can run
```python
rank = 3 # for example
size = 10
dataset = nlp.load_dataset('wikitext', 'wikitext-2-raw-v1', split=f"train[{rank*10}%:{(rank+1)*10}%]")
```
However, this breaks down if you have a number of ranks that doesn't divide cleanly into 100, such as 64 ranks. Is there interest in adding a method shard() that looks like this?
```python
rank = 3
size = 64
dataset = nlp.load_dataset("wikitext", "wikitext-2-raw-v1", split="train").shard(rank=rank, size=size)
```
TensorFlow has a similar API: https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shard. I'd be happy to contribute this code.
Hi Jared,
Interesting, thanks for raising this question. You can also do that after loading with `dataset.select()` or `dataset.filter()` which let you keep only a specific subset of rows in a dataset.
What is your use-case for sharding? |
https://github.com/huggingface/datasets/issues/312 | [Feature request] Add `shard()` method to dataset | Thanks for the pointer to those functions! It's still a little more verbose since you have to manually calculate which ids each rank would keep, but definitely works.
My use case is multi-node, multi-GPU training and avoiding global batches of duplicate elements. I'm using horovod. You can shuffle indices, or set random seeds, but explicitly sharding the dataset up front is the safest and clearest way I've found to do so. | Currently, to shard a dataset into 10 pieces on different ranks, you can run
```python
rank = 3 # for example
size = 10
dataset = nlp.load_dataset('wikitext', 'wikitext-2-raw-v1', split=f"train[{rank*10}%:{(rank+1)*10}%]")
```
However, this breaks down if you have a number of ranks that doesn't divide cleanly into 100, such as 64 ranks. Is there interest in adding a method shard() that looks like this?
```python
rank = 3
size = 64
dataset = nlp.load_dataset("wikitext", "wikitext-2-raw-v1", split="train").shard(rank=rank, size=size)
```
TensorFlow has a similar API: https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shard. I'd be happy to contribute this code. | 71 | [Feature request] Add `shard()` method to dataset
Currently, to shard a dataset into 10 pieces on different ranks, you can run
```python
rank = 3 # for example
size = 10
dataset = nlp.load_dataset('wikitext', 'wikitext-2-raw-v1', split=f"train[{rank*10}%:{(rank+1)*10}%]")
```
However, this breaks down if you have a number of ranks that doesn't divide cleanly into 100, such as 64 ranks. Is there interest in adding a method shard() that looks like this?
```python
rank = 3
size = 64
dataset = nlp.load_dataset("wikitext", "wikitext-2-raw-v1", split="train").shard(rank=rank, size=size)
```
TensorFlow has a similar API: https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shard. I'd be happy to contribute this code.
Thanks for the pointer to those functions! It's still a little more verbose since you have to manually calculate which ids each rank would keep, but definitely works.
My use case is multi-node, multi-GPU training and avoiding global batches of duplicate elements. I'm using horovod. You can shuffle indices, or set random seeds, but explicitly sharding the dataset up front is the safest and clearest way I've found to do so. |
https://github.com/huggingface/datasets/issues/302 | Question - Sign Language Datasets | Even more complicating -
As I see it, datasets can have "addons".
For example, the WebNLG dataset is a dataset for data-to-text. However, a work of mine and other works enriched this dataset with text plans / underlying text structures. In that case, I see a need to load the dataset "WebNLG" with "plans" addon.
Same for sign language - if there is a dataset of videos, one addon can be to run OpenPose, another to run ARKit4 pose estimation, and another to run PoseNet, or even just a video embedding addon. (which are expensive to run individually for everyone who wants to use these data)
This is something I dabbled with my own implementation to a [research datasets library](https://github.com/AmitMY/meta-scholar/) and I love to get the discussion going on these topics. | An emerging field in NLP is SLP - sign language processing.
I was wondering about adding datasets here, specifically because it's shaping up to be large and easily usable.
The metrics for sign language to text translation are the same.
So, what do you think about (me, or others) adding datasets here?
An example dataset would be [RWTH-PHOENIX-Weather 2014 T](https://www-i6.informatik.rwth-aachen.de/~koller/RWTH-PHOENIX-2014-T/)
For every item in the dataset, the data object includes:
1. video_path - path to mp4 file
2. pose_path - a path to `.pose` file with human pose landmarks
3. openpose_path - a path to a `.json` file with human pose landmarks
4. gloss - string
5. text - string
6. video_metadata - height, width, frames, framerate
------
To make it a tad more complicated - what if sign language libraries add requirements to `nlp`? for example, sign language is commonly annotated using `ilex`, `eaf`, or `srt` files, which are all loadable as text, but there is no reason for the dataset to parse that file by itself, if libraries exist to do so. | 131 | Question - Sign Language Datasets
An emerging field in NLP is SLP - sign language processing.
I was wondering about adding datasets here, specifically because it's shaping up to be large and easily usable.
The metrics for sign language to text translation are the same.
So, what do you think about (me, or others) adding datasets here?
An example dataset would be [RWTH-PHOENIX-Weather 2014 T](https://www-i6.informatik.rwth-aachen.de/~koller/RWTH-PHOENIX-2014-T/)
For every item in the dataset, the data object includes:
1. video_path - path to mp4 file
2. pose_path - a path to `.pose` file with human pose landmarks
3. openpose_path - a path to a `.json` file with human pose landmarks
4. gloss - string
5. text - string
6. video_metadata - height, width, frames, framerate
------
To make it a tad more complicated - what if sign language libraries add requirements to `nlp`? for example, sign language is commonly annotated using `ilex`, `eaf`, or `srt` files, which are all loadable as text, but there is no reason for the dataset to parse that file by itself, if libraries exist to do so.
Even more complicating -
As I see it, datasets can have "addons".
For example, the WebNLG dataset is a dataset for data-to-text. However, a work of mine and other works enriched this dataset with text plans / underlying text structures. In that case, I see a need to load the dataset "WebNLG" with "plans" addon.
Same for sign language - if there is a dataset of videos, one addon can be to run OpenPose, another to run ARKit4 pose estimation, and another to run PoseNet, or even just a video embedding addon. (which are expensive to run individually for everyone who wants to use these data)
This is something I dabbled with my own implementation to a [research datasets library](https://github.com/AmitMY/meta-scholar/) and I love to get the discussion going on these topics. |
https://github.com/huggingface/datasets/issues/302 | Question - Sign Language Datasets | This is a really cool idea !
The example for data objects you gave for the RWTH-PHOENIX-Weather 2014 T dataset can totally fit inside the library.
For your point about formats like `ilex`, `eaf`, or `srt`, it is possible to use any library in your dataset script.
However most user probably won't need these libraries, as most datasets don't need them, and therefore it's unlikely that we will have them in the minimum requirements to use `nlp` (we want to keep it as light-weight as possible). If a user wants to load your dataset and doesn't have the libraries you need, an error is raised asking the user to install them.
More generally, we plan to have something like a `requirements.txt` per dataset. This could also be a place for addons as you said. What do you think ? | An emerging field in NLP is SLP - sign language processing.
I was wondering about adding datasets here, specifically because it's shaping up to be large and easily usable.
The metrics for sign language to text translation are the same.
So, what do you think about (me, or others) adding datasets here?
An example dataset would be [RWTH-PHOENIX-Weather 2014 T](https://www-i6.informatik.rwth-aachen.de/~koller/RWTH-PHOENIX-2014-T/)
For every item in the dataset, the data object includes:
1. video_path - path to mp4 file
2. pose_path - a path to `.pose` file with human pose landmarks
3. openpose_path - a path to a `.json` file with human pose landmarks
4. gloss - string
5. text - string
6. video_metadata - height, width, frames, framerate
------
To make it a tad more complicated - what if sign language libraries add requirements to `nlp`? for example, sign language is commonly annotated using `ilex`, `eaf`, or `srt` files, which are all loadable as text, but there is no reason for the dataset to parse that file by itself, if libraries exist to do so. | 139 | Question - Sign Language Datasets
An emerging field in NLP is SLP - sign language processing.
I was wondering about adding datasets here, specifically because it's shaping up to be large and easily usable.
The metrics for sign language to text translation are the same.
So, what do you think about (me, or others) adding datasets here?
An example dataset would be [RWTH-PHOENIX-Weather 2014 T](https://www-i6.informatik.rwth-aachen.de/~koller/RWTH-PHOENIX-2014-T/)
For every item in the dataset, the data object includes:
1. video_path - path to mp4 file
2. pose_path - a path to `.pose` file with human pose landmarks
3. openpose_path - a path to a `.json` file with human pose landmarks
4. gloss - string
5. text - string
6. video_metadata - height, width, frames, framerate
------
To make it a tad more complicated - what if sign language libraries add requirements to `nlp`? for example, sign language is commonly annotated using `ilex`, `eaf`, or `srt` files, which are all loadable as text, but there is no reason for the dataset to parse that file by itself, if libraries exist to do so.
This is a really cool idea !
The example for data objects you gave for the RWTH-PHOENIX-Weather 2014 T dataset can totally fit inside the library.
For your point about formats like `ilex`, `eaf`, or `srt`, it is possible to use any library in your dataset script.
However most user probably won't need these libraries, as most datasets don't need them, and therefore it's unlikely that we will have them in the minimum requirements to use `nlp` (we want to keep it as light-weight as possible). If a user wants to load your dataset and doesn't have the libraries you need, an error is raised asking the user to install them.
More generally, we plan to have something like a `requirements.txt` per dataset. This could also be a place for addons as you said. What do you think ? |
https://github.com/huggingface/datasets/issues/302 | Question - Sign Language Datasets | Thanks, Quentin, I think a `requirements.txt` per dataset will be a good thing.
I will work on adding this dataset next week, and once we sort all of the kinks, I'll add more. | An emerging field in NLP is SLP - sign language processing.
I was wondering about adding datasets here, specifically because it's shaping up to be large and easily usable.
The metrics for sign language to text translation are the same.
So, what do you think about (me, or others) adding datasets here?
An example dataset would be [RWTH-PHOENIX-Weather 2014 T](https://www-i6.informatik.rwth-aachen.de/~koller/RWTH-PHOENIX-2014-T/)
For every item in the dataset, the data object includes:
1. video_path - path to mp4 file
2. pose_path - a path to `.pose` file with human pose landmarks
3. openpose_path - a path to a `.json` file with human pose landmarks
4. gloss - string
5. text - string
6. video_metadata - height, width, frames, framerate
------
To make it a tad more complicated - what if sign language libraries add requirements to `nlp`? for example, sign language is commonly annotated using `ilex`, `eaf`, or `srt` files, which are all loadable as text, but there is no reason for the dataset to parse that file by itself, if libraries exist to do so. | 33 | Question - Sign Language Datasets
An emerging field in NLP is SLP - sign language processing.
I was wondering about adding datasets here, specifically because it's shaping up to be large and easily usable.
The metrics for sign language to text translation are the same.
So, what do you think about (me, or others) adding datasets here?
An example dataset would be [RWTH-PHOENIX-Weather 2014 T](https://www-i6.informatik.rwth-aachen.de/~koller/RWTH-PHOENIX-2014-T/)
For every item in the dataset, the data object includes:
1. video_path - path to mp4 file
2. pose_path - a path to `.pose` file with human pose landmarks
3. openpose_path - a path to a `.json` file with human pose landmarks
4. gloss - string
5. text - string
6. video_metadata - height, width, frames, framerate
------
To make it a tad more complicated - what if sign language libraries add requirements to `nlp`? for example, sign language is commonly annotated using `ilex`, `eaf`, or `srt` files, which are all loadable as text, but there is no reason for the dataset to parse that file by itself, if libraries exist to do so.
Thanks, Quentin, I think a `requirements.txt` per dataset will be a good thing.
I will work on adding this dataset next week, and once we sort all of the kinks, I'll add more. |
https://github.com/huggingface/datasets/issues/301 | Setting cache_dir gives error on wikipedia download | Whoops didn't mean to close this one.
I did some changes, could you try to run it from the master branch ? | First of all thank you for a super handy library! I'd like to download large files to a specific drive so I set `cache_dir=my_path`. This works fine with e.g. imdb and squad. But on wikipedia I get an error:
```
nlp.load_dataset('wikipedia', '20200501.de', split = 'train', cache_dir=my_path)
```
```
OSError Traceback (most recent call last)
<ipython-input-2-23551344d7bc> in <module>
1 import nlp
----> 2 nlp.load_dataset('wikipedia', '20200501.de', split = 'train', cache_dir=path)
~/anaconda3/envs/fastai2/lib/python3.7/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/anaconda3/envs/fastai2/lib/python3.7/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
385 with utils.temporary_assignment(self, "_cache_dir", tmp_data_dir):
386 reader = ArrowReader(self._cache_dir, self.info)
--> 387 reader.download_from_hf_gcs(self._cache_dir, self._relative_data_dir(with_version=True))
388 downloaded_info = DatasetInfo.from_directory(self._cache_dir)
389 self.info.update(downloaded_info)
~/anaconda3/envs/fastai2/lib/python3.7/site-packages/nlp/arrow_reader.py in download_from_hf_gcs(self, cache_dir, relative_data_dir)
231 remote_dataset_info = os.path.join(remote_cache_dir, "dataset_info.json")
232 downloaded_dataset_info = cached_path(remote_dataset_info)
--> 233 os.rename(downloaded_dataset_info, os.path.join(cache_dir, "dataset_info.json"))
234 if self._info is not None:
235 self._info.update(self._info.from_directory(cache_dir))
OSError: [Errno 18] Invalid cross-device link: '/home/local/NTU/nn/.cache/huggingface/datasets/025fa4fd4f04aaafc9e939260fbc8f0bb190ce14c61310c8ae1ddd1dcb31f88c.9637f367b6711a79ca478be55fe6989b8aea4941b7ef7adc67b89ff403020947' -> '/data/nn/nlp/wikipedia/20200501.de/1.0.0.incomplete/dataset_info.json'
``` | 22 | Setting cache_dir gives error on wikipedia download
First of all thank you for a super handy library! I'd like to download large files to a specific drive so I set `cache_dir=my_path`. This works fine with e.g. imdb and squad. But on wikipedia I get an error:
```
nlp.load_dataset('wikipedia', '20200501.de', split = 'train', cache_dir=my_path)
```
```
OSError Traceback (most recent call last)
<ipython-input-2-23551344d7bc> in <module>
1 import nlp
----> 2 nlp.load_dataset('wikipedia', '20200501.de', split = 'train', cache_dir=path)
~/anaconda3/envs/fastai2/lib/python3.7/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/anaconda3/envs/fastai2/lib/python3.7/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
385 with utils.temporary_assignment(self, "_cache_dir", tmp_data_dir):
386 reader = ArrowReader(self._cache_dir, self.info)
--> 387 reader.download_from_hf_gcs(self._cache_dir, self._relative_data_dir(with_version=True))
388 downloaded_info = DatasetInfo.from_directory(self._cache_dir)
389 self.info.update(downloaded_info)
~/anaconda3/envs/fastai2/lib/python3.7/site-packages/nlp/arrow_reader.py in download_from_hf_gcs(self, cache_dir, relative_data_dir)
231 remote_dataset_info = os.path.join(remote_cache_dir, "dataset_info.json")
232 downloaded_dataset_info = cached_path(remote_dataset_info)
--> 233 os.rename(downloaded_dataset_info, os.path.join(cache_dir, "dataset_info.json"))
234 if self._info is not None:
235 self._info.update(self._info.from_directory(cache_dir))
OSError: [Errno 18] Invalid cross-device link: '/home/local/NTU/nn/.cache/huggingface/datasets/025fa4fd4f04aaafc9e939260fbc8f0bb190ce14c61310c8ae1ddd1dcb31f88c.9637f367b6711a79ca478be55fe6989b8aea4941b7ef7adc67b89ff403020947' -> '/data/nn/nlp/wikipedia/20200501.de/1.0.0.incomplete/dataset_info.json'
```
Whoops didn't mean to close this one.
I did some changes, could you try to run it from the master branch ? |
https://github.com/huggingface/datasets/issues/297 | Error in Demo for Specific Datasets | Thanks for reporting these errors :)
I can actually see two issues here.
First, datasets like `natural_questions` require apache_beam to be processed. Right now the import is not at the right place so we have this error message. However, even the imports are fixed, the nlp viewer doesn't actually have the resources to process NQ right now so we'll have to wait until we have a version that we've already processed on our google storage (that's what we've done for wikipedia for example).
Second, datasets like `newsroom` require manual downloads as we're not allowed to redistribute the data ourselves (if I'm not wrong). An error message should be displayed saying that we're not allowed to show the dataset.
I can fix the first issue with the imports but for the second one I think we'll have to see with @srush to show a message for datasets that require manual downloads (it can be checked whether a dataset requires manual downloads if `dataset_builder_instance.manual_download_instructions is not None`).
| Selecting `natural_questions` or `newsroom` dataset in the online demo results in an error similar to the following.
| 165 | Error in Demo for Specific Datasets
Selecting `natural_questions` or `newsroom` dataset in the online demo results in an error similar to the following.
Thanks for reporting these errors :)
I can actually see two issues here.
First, datasets like `natural_questions` require apache_beam to be processed. Right now the import is not at the right place so we have this error message. However, even the imports are fixed, the nlp viewer doesn't actually have the resources to process NQ right now so we'll have to wait until we have a version that we've already processed on our google storage (that's what we've done for wikipedia for example).
Second, datasets like `newsroom` require manual downloads as we're not allowed to redistribute the data ourselves (if I'm not wrong). An error message should be displayed saying that we're not allowed to show the dataset.
I can fix the first issue with the imports but for the second one I think we'll have to see with @srush to show a message for datasets that require manual downloads (it can be checked whether a dataset requires manual downloads if `dataset_builder_instance.manual_download_instructions is not None`).
|
https://github.com/huggingface/datasets/issues/297 | Error in Demo for Specific Datasets | We don't plan to host the source files of newsroom ourselves for now.
You can still get the dataset if you follow the download instructions given by `dataset = load_dataset('newsroom')` though.
The viewer also shows the instructions now.
Closing this one. If you have other questions, feel free to re-open :) | Selecting `natural_questions` or `newsroom` dataset in the online demo results in an error similar to the following.
| 51 | Error in Demo for Specific Datasets
Selecting `natural_questions` or `newsroom` dataset in the online demo results in an error similar to the following.
We don't plan to host the source files of newsroom ourselves for now.
You can still get the dataset if you follow the download instructions given by `dataset = load_dataset('newsroom')` though.
The viewer also shows the instructions now.
Closing this one. If you have other questions, feel free to re-open :) |
https://github.com/huggingface/datasets/issues/296 | snli -1 labels | @jxmorris12 , we use `-1` to label examples for which `gold label` is missing (`gold label = -` in the original dataset). | I'm trying to train a model on the SNLI dataset. Why does it have so many -1 labels?
```
import nlp
from collections import Counter
data = nlp.load_dataset('snli')['train']
print(Counter(data['label']))
Counter({0: 183416, 2: 183187, 1: 182764, -1: 785})
```
| 22 | snli -1 labels
I'm trying to train a model on the SNLI dataset. Why does it have so many -1 labels?
```
import nlp
from collections import Counter
data = nlp.load_dataset('snli')['train']
print(Counter(data['label']))
Counter({0: 183416, 2: 183187, 1: 182764, -1: 785})
```
@jxmorris12 , we use `-1` to label examples for which `gold label` is missing (`gold label = -` in the original dataset). |
https://github.com/huggingface/datasets/issues/296 | snli -1 labels | Thanks @mariamabarham! so the original dataset is missing some labels? That is weird. Is standard practice just to discard those examples training/eval? | I'm trying to train a model on the SNLI dataset. Why does it have so many -1 labels?
```
import nlp
from collections import Counter
data = nlp.load_dataset('snli')['train']
print(Counter(data['label']))
Counter({0: 183416, 2: 183187, 1: 182764, -1: 785})
```
| 22 | snli -1 labels
I'm trying to train a model on the SNLI dataset. Why does it have so many -1 labels?
```
import nlp
from collections import Counter
data = nlp.load_dataset('snli')['train']
print(Counter(data['label']))
Counter({0: 183416, 2: 183187, 1: 182764, -1: 785})
```
Thanks @mariamabarham! so the original dataset is missing some labels? That is weird. Is standard practice just to discard those examples training/eval? |
https://github.com/huggingface/datasets/issues/296 | snli -1 labels | Yes the original dataset is missing some labels maybe @sleepinyourhat , @gangeli can correct me if I'm wrong
For my personal opinion at least if you want your model to learn to predict no answer (-1) you can leave it their but otherwise you can discard them. | I'm trying to train a model on the SNLI dataset. Why does it have so many -1 labels?
```
import nlp
from collections import Counter
data = nlp.load_dataset('snli')['train']
print(Counter(data['label']))
Counter({0: 183416, 2: 183187, 1: 182764, -1: 785})
```
| 47 | snli -1 labels
I'm trying to train a model on the SNLI dataset. Why does it have so many -1 labels?
```
import nlp
from collections import Counter
data = nlp.load_dataset('snli')['train']
print(Counter(data['label']))
Counter({0: 183416, 2: 183187, 1: 182764, -1: 785})
```
Yes the original dataset is missing some labels maybe @sleepinyourhat , @gangeli can correct me if I'm wrong
For my personal opinion at least if you want your model to learn to predict no answer (-1) you can leave it their but otherwise you can discard them. |
https://github.com/huggingface/datasets/issues/294 | Cannot load arxiv dataset on MacOS? | I couldn't replicate this issue on my macbook :/
Could you try to play with different encodings in `with open(path, encoding=...) as f` in scientific_papers.py:L108 ? | I am having trouble loading the `"arxiv"` config from the `"scientific_papers"` dataset on MacOS. When I try loading the dataset with:
```python
arxiv = nlp.load_dataset("scientific_papers", "arxiv")
```
I get the following stack trace:
```bash
JSONDecodeError Traceback (most recent call last)
<ipython-input-2-8e00c55d5a59> in <module>
----> 1 arxiv = nlp.load_dataset("scientific_papers", "arxiv")
~/miniconda3/envs/t2t/lib/python3.7/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/miniconda3/envs/t2t/lib/python3.7/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/miniconda3/envs/t2t/lib/python3.7/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/miniconda3/envs/t2t/lib/python3.7/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
662
663 generator = self._generate_examples(**split_generator.gen_kwargs)
--> 664 for key, record in utils.tqdm(generator, unit=" examples", total=split_info.num_examples, leave=False):
665 example = self.info.features.encode_example(record)
666 writer.write(example)
~/miniconda3/envs/t2t/lib/python3.7/site-packages/tqdm/std.py in __iter__(self)
1106 fp_write=getattr(self.fp, 'write', sys.stderr.write))
1107
-> 1108 for obj in iterable:
1109 yield obj
1110 # Update and possibly print the progressbar.
~/miniconda3/envs/t2t/lib/python3.7/site-packages/nlp/datasets/scientific_papers/107a416c0e1958cb846f5934b5aae292f7884a5b27e86af3f3ef1a093e058bbc/scientific_papers.py in _generate_examples(self, path)
114 # "section_names": list[str], list of section names.
115 # "sections": list[list[str]], list of sections (list of paragraphs)
--> 116 d = json.loads(line)
117 summary = "\n".join(d["abstract_text"])
118 # In original paper, <S> and </S> are not used in vocab during training
~/miniconda3/envs/t2t/lib/python3.7/json/__init__.py in loads(s, encoding, cls, object_hook, parse_float, parse_int, parse_constant, object_pairs_hook, **kw)
346 parse_int is None and parse_float is None and
347 parse_constant is None and object_pairs_hook is None and not kw):
--> 348 return _default_decoder.decode(s)
349 if cls is None:
350 cls = JSONDecoder
~/miniconda3/envs/t2t/lib/python3.7/json/decoder.py in decode(self, s, _w)
335
336 """
--> 337 obj, end = self.raw_decode(s, idx=_w(s, 0).end())
338 end = _w(s, end).end()
339 if end != len(s):
~/miniconda3/envs/t2t/lib/python3.7/json/decoder.py in raw_decode(self, s, idx)
351 """
352 try:
--> 353 obj, end = self.scan_once(s, idx)
354 except StopIteration as err:
355 raise JSONDecodeError("Expecting value", s, err.value) from None
JSONDecodeError: Unterminated string starting at: line 1 column 46983 (char 46982)
163502 examples [02:10, 2710.68 examples/s]
```
I am not sure how to trace back to the specific JSON file that has the "Unterminated string". Also, I do not get this error on colab so I suspect it may be MacOS specific. Copy pasting the relevant lines from `transformers-cli env` below:
- Platform: Darwin-19.5.0-x86_64-i386-64bit
- Python version: 3.7.5
- PyTorch version (GPU?): 1.5.0 (False)
- Tensorflow version (GPU?): 2.2.0 (False)
Any ideas? | 26 | Cannot load arxiv dataset on MacOS?
I am having trouble loading the `"arxiv"` config from the `"scientific_papers"` dataset on MacOS. When I try loading the dataset with:
```python
arxiv = nlp.load_dataset("scientific_papers", "arxiv")
```
I get the following stack trace:
```bash
JSONDecodeError Traceback (most recent call last)
<ipython-input-2-8e00c55d5a59> in <module>
----> 1 arxiv = nlp.load_dataset("scientific_papers", "arxiv")
~/miniconda3/envs/t2t/lib/python3.7/site-packages/nlp/load.py in load_dataset(path, name, version, data_dir, data_files, split, cache_dir, download_config, download_mode, ignore_verifications, save_infos, **config_kwargs)
522 download_mode=download_mode,
523 ignore_verifications=ignore_verifications,
--> 524 save_infos=save_infos,
525 )
526
~/miniconda3/envs/t2t/lib/python3.7/site-packages/nlp/builder.py in download_and_prepare(self, download_config, download_mode, ignore_verifications, save_infos, try_from_hf_gcs, dl_manager, **download_and_prepare_kwargs)
430 verify_infos = not save_infos and not ignore_verifications
431 self._download_and_prepare(
--> 432 dl_manager=dl_manager, verify_infos=verify_infos, **download_and_prepare_kwargs
433 )
434 # Sync info
~/miniconda3/envs/t2t/lib/python3.7/site-packages/nlp/builder.py in _download_and_prepare(self, dl_manager, verify_infos, **prepare_split_kwargs)
481 try:
482 # Prepare split will record examples associated to the split
--> 483 self._prepare_split(split_generator, **prepare_split_kwargs)
484 except OSError:
485 raise OSError("Cannot find data file. " + (self.manual_download_instructions or ""))
~/miniconda3/envs/t2t/lib/python3.7/site-packages/nlp/builder.py in _prepare_split(self, split_generator)
662
663 generator = self._generate_examples(**split_generator.gen_kwargs)
--> 664 for key, record in utils.tqdm(generator, unit=" examples", total=split_info.num_examples, leave=False):
665 example = self.info.features.encode_example(record)
666 writer.write(example)
~/miniconda3/envs/t2t/lib/python3.7/site-packages/tqdm/std.py in __iter__(self)
1106 fp_write=getattr(self.fp, 'write', sys.stderr.write))
1107
-> 1108 for obj in iterable:
1109 yield obj
1110 # Update and possibly print the progressbar.
~/miniconda3/envs/t2t/lib/python3.7/site-packages/nlp/datasets/scientific_papers/107a416c0e1958cb846f5934b5aae292f7884a5b27e86af3f3ef1a093e058bbc/scientific_papers.py in _generate_examples(self, path)
114 # "section_names": list[str], list of section names.
115 # "sections": list[list[str]], list of sections (list of paragraphs)
--> 116 d = json.loads(line)
117 summary = "\n".join(d["abstract_text"])
118 # In original paper, <S> and </S> are not used in vocab during training
~/miniconda3/envs/t2t/lib/python3.7/json/__init__.py in loads(s, encoding, cls, object_hook, parse_float, parse_int, parse_constant, object_pairs_hook, **kw)
346 parse_int is None and parse_float is None and
347 parse_constant is None and object_pairs_hook is None and not kw):
--> 348 return _default_decoder.decode(s)
349 if cls is None:
350 cls = JSONDecoder
~/miniconda3/envs/t2t/lib/python3.7/json/decoder.py in decode(self, s, _w)
335
336 """
--> 337 obj, end = self.raw_decode(s, idx=_w(s, 0).end())
338 end = _w(s, end).end()
339 if end != len(s):
~/miniconda3/envs/t2t/lib/python3.7/json/decoder.py in raw_decode(self, s, idx)
351 """
352 try:
--> 353 obj, end = self.scan_once(s, idx)
354 except StopIteration as err:
355 raise JSONDecodeError("Expecting value", s, err.value) from None
JSONDecodeError: Unterminated string starting at: line 1 column 46983 (char 46982)
163502 examples [02:10, 2710.68 examples/s]
```
I am not sure how to trace back to the specific JSON file that has the "Unterminated string". Also, I do not get this error on colab so I suspect it may be MacOS specific. Copy pasting the relevant lines from `transformers-cli env` below:
- Platform: Darwin-19.5.0-x86_64-i386-64bit
- Python version: 3.7.5
- PyTorch version (GPU?): 1.5.0 (False)
- Tensorflow version (GPU?): 2.2.0 (False)
Any ideas?
I couldn't replicate this issue on my macbook :/
Could you try to play with different encodings in `with open(path, encoding=...) as f` in scientific_papers.py:L108 ? |