Convert dataset to Parquet (#5)

- Convert dataset to Parquet (ce6b283add5660041bdec82d37ebc55c91af9025)
- Add compguesswhat-zero_shot data files (2e4aa4c52a868a558bee4bb0dadf6e8f7d9213bb)
- Delete loading script (b4a856ff9d26bcd14f51f4e84bbcc1e759d55858)
- Delete loading script auxiliary file (d80f88a46da2c2b877427dc88df535fec22670d1)

README.md

@@ -1,15 +1,14 @@
 ---
 annotations_creators:
 - machine-generated
-language:
-- en
 language_creators:
 - found
+language:
+- en
 license:
 - unknown
 multilinguality:
 - monolingual
-pretty_name: CompGuessWhat?!
 size_categories:
 - 100K<n<1M
 source_datasets:
@@ -19,6 +18,7 @@ task_categories:
 task_ids:
 - visual-question-answering
 paperswithcode_id: compguesswhat
+pretty_name: CompGuessWhat?!
 dataset_info:
 - config_name: compguesswhat-original
   features:
@@ -84,16 +84,16 @@ dataset_info:
         sequence: float32
   splits:
   - name: train
-    num_bytes: 126548689
+    num_bytes: 123556580
     num_examples: 46341
   - name: validation
-    num_bytes: 26055261
+    num_bytes: 25441428
     num_examples: 9738
   - name: test
-    num_bytes: 25981593
+    num_bytes: 25369227
     num_examples: 9621
-  download_size: 107201655
-  dataset_size: 178585543
+  download_size: 105349759
+  dataset_size: 174367235
 - config_name: compguesswhat-zero_shot
   features:
   - name: id
@@ -159,19 +159,38 @@ dataset_info:
         dtype: string
   splits:
   - name: nd_valid
-    num_bytes: 13557059
+    num_bytes: 13510589
     num_examples: 5343
   - name: nd_test
-    num_bytes: 36352201
+    num_bytes: 36228021
     num_examples: 13836
   - name: od_valid
-    num_bytes: 14093233
+    num_bytes: 14051972
     num_examples: 5372
   - name: od_test
-    num_bytes: 33049755
+    num_bytes: 32950869
     num_examples: 13300
-  download_size: 4845966
-  dataset_size: 97052248
+  download_size: 6548812
+  dataset_size: 96741451
+configs:
+- config_name: compguesswhat-original
+  data_files:
+  - split: train
+    path: compguesswhat-original/train-*
+  - split: validation
+    path: compguesswhat-original/validation-*
+  - split: test
+    path: compguesswhat-original/test-*
+- config_name: compguesswhat-zero_shot
+  data_files:
+  - split: nd_valid
+    path: compguesswhat-zero_shot/nd_valid-*
+  - split: nd_test
+    path: compguesswhat-zero_shot/nd_test-*
+  - split: od_valid
+    path: compguesswhat-zero_shot/od_valid-*
+  - split: od_test
+    path: compguesswhat-zero_shot/od_test-*
 ---
 
 # Dataset Card for "compguesswhat"

compguesswhat-original/test-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16b2a80dd58e66aad5fe8857c95a812e2206ae05a55ec6cfa89d38518577415a
+size 15268653

compguesswhat-original/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4099d3ab1826c0acbf47710ffa8197cf52204cfaa6df44149c1d60ee01dd41f
+size 74763809

compguesswhat-original/validation-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:024d592321b0bc7f7f8d6419b1f45f8e188875f4bf187a59037c694b6e806784
+size 15317297

compguesswhat-zero_shot/nd_test-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7bfb006600e91cf151b644b1495c5688268997cd024c9ce832d7500302d690e
+size 2434268

compguesswhat-zero_shot/nd_valid-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03d7d02f63bac35c043d2e73f1cceec10d365440f165348ff86dc8cf3fc9fcd4
+size 929562

compguesswhat-zero_shot/od_test-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67eee5a8432934369fa4d4eb33e7fd11e4b57fb85bd2d5e826e46905b7d1a54b
+size 2255581

compguesswhat-zero_shot/od_valid-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e6b4ef094118fb380e70a83dbcc2fa1c43cc56f655f48d419b8bfbd72bf795ac
+size 929401

compguesswhat.py

@@ -1,254 +0,0 @@
-import gzip
-import json
-import os
-
-import datasets
-
-
-class CompguesswhatConfig(datasets.BuilderConfig):
-    """BuilderConfig for CompGuessWhat?!"""
-
-    def __init__(self, data_url, splits, gameplay_scenario, **kwargs):
-        """
-
-        Args:
-            gameplay_scenario: to specify if we want to load original CompGuessWhat?! split ('original') or
-            the zero-shot reference games based on NOCAPS images ('zero_shot')
-            **kwargs: keyword arguments forwarded to super.
-        """
-        super(CompguesswhatConfig, self).__init__(
-            version=datasets.Version("0.2.0", "Second CompGuessWhat?! release"), **kwargs
-        )
-        assert gameplay_scenario in (
-            "original",
-            "zero_shot",
-        ), "Invalid choice for parameter 'gameplay_scenario': {gameplay_scenario}. Valid values are ('original', 'zero_shot')."
-
-        self.gameplay_scenario = gameplay_scenario
-        self.splits = splits
-        self.data_url = data_url
-
-
-class Compguesswhat(datasets.GeneratorBasedBuilder):
-    _CITATION = """\
-        @inproceedings{suglia2020compguesswhat,
-          title={CompGuessWhat?!: a Multi-task Evaluation Framework for Grounded Language Learning},
-          author={Suglia, Alessandro, Konstas, Ioannis, Vanzo, Andrea, Bastianelli, Emanuele, Desmond Elliott, Stella Frank and Oliver Lemon},
-          booktitle={Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics},
-          year={2020}
-        }
-    """
-
-    _DESCRIPTION = """
-        CompGuessWhat?! is an instance of a multi-task framework for evaluating the quality of learned neural representations,
-        in particular concerning attribute grounding. Use this dataset if you want to use the set of games whose reference
-        scene is an image in VisualGenome. Visit the website for more details: https://compguesswhat.github.io
-    """
-
-    BUILDER_CONFIGS = [
-        CompguesswhatConfig(
-            name="compguesswhat-original",
-            gameplay_scenario="original",
-            description="CompGuessWhat?! subset of games from the original GuessWhat?! dataset",
-            data_url="https://www.dropbox.com/s/qd9wlydpkpmq8rr/compguesswhat-original.zip?dl=1",
-            splits={
-                "train": "compguesswhat.train.jsonl.gz",
-                "valid": "compguesswhat.valid.jsonl.gz",
-                "test": "compguesswhat.test.jsonl.gz",
-            },
-        ),
-        CompguesswhatConfig(
-            name="compguesswhat-zero_shot",
-            gameplay_scenario="zero_shot",
-            description="CompGuessWhat?! reference set of games for zero-shot evaluation using NOCAPS images",
-            data_url="https://www.dropbox.com/s/f5o2t7aiok7kpcm/compguesswhat-zero_shot.zip?dl=1",
-            splits={
-                "nd_valid": "compguesswhat.nd_valid.jsonl.gz",
-                "nd_test": "compguesswhat.nd_test.jsonl.gz",
-                "od_valid": "compguesswhat.od_valid.jsonl.gz",
-                "od_test": "compguesswhat.od_test.jsonl.gz",
-            },
-        ),
-    ]
-
-    VERSION = datasets.Version("0.2.0")
-
-    def _info(self):
-        if self.config.gameplay_scenario == "original":
-            return datasets.DatasetInfo(
-                # This is the description that will appear on the datasets page.
-                description=self._DESCRIPTION,
-                # datasets.features.FeatureConnectors
-                features=datasets.Features(
-                    {
-                        "id": datasets.Value("int32"),
-                        "target_id": datasets.Value("int32"),
-                        "timestamp": datasets.Value("string"),
-                        "status": datasets.Value("string"),
-                        "image": {
-                            # this is the image ID in GuessWhat?! which corresponds to the MSCOCO id
-                            "id": datasets.Value("int32"),
-                            "file_name": datasets.Value("string"),
-                            "flickr_url": datasets.Value("string"),
-                            "coco_url": datasets.Value("string"),
-                            "height": datasets.Value("int32"),
-                            "width": datasets.Value("int32"),
-                            # this field represents the corresponding image metadata that can be found in VisualGenome
-                            # in the file image_data.json
-                            # We copy it over so that we avoid any confusion or possible wrong URL
-                            # Please use the original image files to resolve photos
-                            "visual_genome": {
-                                "width": datasets.Value("int32"),
-                                "height": datasets.Value("int32"),
-                                "url": datasets.Value("string"),
-                                "coco_id": datasets.Value("int32"),
-                                # this is the actual VisualGenome image ID
-                                # because we can't rely store it as an integer we same it as string
-                                "flickr_id": datasets.Value("string"),
-                                "image_id": datasets.Value("string"),
-                            },
-                        },
-                        "qas": datasets.features.Sequence(
-                            {
-                                "question": datasets.Value("string"),
-                                "answer": datasets.Value("string"),
-                                "id": datasets.Value("int32"),
-                            }
-                        ),
-                        "objects": datasets.features.Sequence(
-                            {
-                                "id": datasets.Value("int32"),
-                                "bbox": datasets.Sequence(datasets.Value("float32"), length=4),
-                                "category": datasets.Value("string"),
-                                "area": datasets.Value("float32"),
-                                "category_id": datasets.Value("int32"),
-                                "segment": datasets.features.Sequence(
-                                    datasets.features.Sequence(datasets.Value("float32"))
-                                ),
-                            }
-                        ),
-                    }
-                ),
-                # If there's a common (input, target) tuple from the features,
-                # specify them here. They'll be used if as_supervised=True in
-                # builder.as_dataset.
-                supervised_keys=None,
-                # Homepage of the dataset for documentation
-                homepage="https://compguesswhat.github.io/",
-                citation=self._CITATION,
-            )
-        elif self.config.gameplay_scenario == "zero_shot":
-            return datasets.DatasetInfo(
-                # This is the description that will appear on the datasets page.
-                description=self._DESCRIPTION,
-                # datasets.features.FeatureConnectors
-                features=datasets.Features(
-                    {
-                        "id": datasets.Value("int32"),
-                        "target_id": datasets.Value("string"),
-                        "status": datasets.Value("string"),
-                        "image": {
-                            "id": datasets.Value("int32"),
-                            "file_name": datasets.Value("string"),
-                            "coco_url": datasets.Value("string"),
-                            "height": datasets.Value("int32"),
-                            "width": datasets.Value("int32"),
-                            "license": datasets.Value("int32"),
-                            "open_images_id": datasets.Value("string"),
-                            "date_captured": datasets.Value("string"),
-                        },
-                        "objects": datasets.features.Sequence(
-                            {
-                                "id": datasets.Value("string"),
-                                "bbox": datasets.Sequence(datasets.Value("float32"), length=4),
-                                "category": datasets.Value("string"),
-                                "area": datasets.Value("float32"),
-                                "category_id": datasets.Value("int32"),
-                                "IsOccluded": datasets.Value("int32"),
-                                "IsTruncated": datasets.Value("int32"),
-                                "segment": datasets.features.Sequence(
-                                    {
-                                        "MaskPath": datasets.Value("string"),
-                                        "LabelName": datasets.Value("string"),
-                                        "BoxID": datasets.Value("string"),
-                                        "BoxXMin": datasets.Value("string"),
-                                        "BoxXMax": datasets.Value("string"),
-                                        "BoxYMin": datasets.Value("string"),
-                                        "BoxYMax": datasets.Value("string"),
-                                        "PredictedIoU": datasets.Value("string"),
-                                        "Clicks": datasets.Value("string"),
-                                    }
-                                ),
-                            }
-                        ),
-                    }
-                ),
-                # If there's a common (input, target) tuple from the features,
-                # specify them here. They'll be used if as_supervised=True in
-                # builder.as_dataset.
-                supervised_keys=None,
-                # Homepage of the dataset for documentation
-                homepage="https://compguesswhat.github.io/",
-                citation=self._CITATION,
-            )
-
-    def _split_generators(self, dl_manager):
-        """Returns SplitGenerators."""
-        dl_dir = dl_manager.download_and_extract(self.config.data_url)
-
-        splits_gen = []
-
-        for split_id, split_filename in self.config.splits.items():
-            if self.config.gameplay_scenario == "original":
-                if "train" in split_id:
-                    split_name = datasets.Split.TRAIN
-                elif "valid" in split_id:
-                    split_name = datasets.Split.VALIDATION
-                elif "test" in split_id:
-                    split_name = datasets.Split.TEST
-            else:
-                split_name = datasets.Split(split_id)
-
-            full_split_name = "-".join(["compguesswhat", self.config.gameplay_scenario])
-            splits_gen.append(
-                datasets.SplitGenerator(
-                    name=split_name,
-                    gen_kwargs={
-                        "filepath": os.path.join(
-                            dl_dir,
-                            full_split_name,
-                            self.VERSION.version_str,
-                            split_filename,
-                        )
-                    },
-                )
-            )
-
-        return splits_gen
-
-    def _generate_examples(self, filepath):
-        def _extract_game_tuple(data):
-            data = data.decode("utf-8")
-            game = json.loads(data.strip("\n"))
-
-            # we refactor the data structure a bit to fit with the new version
-            game["target_id"] = game["object_id"]
-            if "object_id" in game:
-                del game["object_id"]
-
-            if "questioner_id" in game:
-                del game["questioner_id"]
-            ###
-
-            if "visual_genome" in game["image"]:
-                # We need to cast it to string so that we avoid issues with int size
-                game["image"]["visual_genome"]["image_id"] = str(game["image"]["visual_genome"]["image_id"])
-                game["image"]["visual_genome"]["flickr_id"] = str(game["image"]["visual_genome"]["flickr_id"])
-
-            return game["id"], game
-
-        """Yields examples."""
-        with open(filepath, "rb") as gzip_file:
-            with gzip.open(gzip_file) as in_file:
-                for data in in_file:
-                    yield _extract_game_tuple(data)

create_dummy_data.py

@@ -1,96 +0,0 @@
-import gzip
-import json
-import logging
-import os
-from argparse import ArgumentParser
-
-
-parser = ArgumentParser()
-parser.add_argument(
-    "-d",
-    "--data_path",
-    type=str,
-    required=True,
-    help="Data path containing the CompGuessWhat?! datasets (files with 'jsonl.gz' extension)",
-)
-parser.add_argument(
-    "--examples",
-    type=int,
-    default=5,
-    help="Number of games to consider in the dummy dataset",
-)
-original_data_files = {
-    "train": "compguesswhat.train.jsonl.gz",
-    "valid": "compguesswhat.valid.jsonl.gz",
-    "test": "compguesswhat.test.jsonl.gz",
-}
-
-zs_data_files = {
-    "nd_valid": "compguesswhat.nd_valid.jsonl.gz",
-    "od_valid": "compguesswhat.od_valid.jsonl.gz",
-    "nd_test": "compguesswhat.nd_test.jsonl.gz",
-    "od_test": "compguesswhat.od_test.jsonl.gz",
-}
-
-COMPGUESSWHAT_ROOT = "datasets/compguesswhat/"
-
-
-def create_dummy_data_for_split(data_path, dataset_name, dataset_version, data_files):
-    full_dataset_name = "-".join(["compguesswhat", dataset_name])
-    dummy_data_path = os.path.join(
-        COMPGUESSWHAT_ROOT,
-        "dummy",
-        full_dataset_name,
-        dataset_version,
-        "dummy_data",
-        full_dataset_name,
-        dataset_version,
-    )
-
-    if not os.path.exists(dummy_data_path):
-        os.makedirs(dummy_data_path)
-
-    for split_name, split_file in data_files.items():
-        logging.info(f"Generating dummy data for split {split_name} (num. examples = {args.examples})")
-
-        split_filepath = os.path.join(data_path, full_dataset_name, dataset_version, split_file)
-        logging.info(f"Reading split file {split_filepath}")
-        with gzip.open(split_filepath) as in_file:
-            dummy_filepath = os.path.join(dummy_data_path, split_file)
-            with gzip.open(dummy_filepath, mode="w") as out_file:
-                for i, line in enumerate(in_file):
-                    if i > args.examples:
-                        break
-
-                    data = json.loads(line.strip())
-                    out_file.write(json.dumps(data).encode("utf-8"))
-                    out_file.write(b"\n")
-
-
-def main(args):
-    # args.data_path is the directory containing the already downloaded dataset files
-    # we assume that the dataset test was successful and we have the file dataset_info.json
-    dataset_info_path = os.path.join(COMPGUESSWHAT_ROOT, "dataset_infos.json")
-
-    if not os.path.exists(dataset_info_path):
-        raise ValueError(
-            "The file 'dataset_info.json' doesn't exists. Make sure that you run the dataset tests via datasets-cli."
-        )
-
-    with open(dataset_info_path, encoding="utf-8") as in_file:
-        dataset_info = json.load(in_file)
-
-    dataset_version = dataset_info["compguesswhat-original"]["version"]["version_str"]
-
-    logging.info(f"Creating dummy data for CompGuessWhat?! {dataset_version}")
-
-    logging.info("Original dataset...")
-    create_dummy_data_for_split(args.data_path, "original", dataset_version, original_data_files)
-
-    logging.info("Zero-shot dataset...")
-    create_dummy_data_for_split(args.data_path, "zero_shot", dataset_version, zs_data_files)
-
-
-if __name__ == "__main__":
-    args = parser.parse_args()
-    main(args)