Datasets:
Sebastian Gehrmann
commited on
Commit
•
5ddf90b
1
Parent(s):
2eeba6b
initial loader
Browse files- dataset_infos.json +1 -1
- viggo.py +1 -1
dataset_infos.json
CHANGED
@@ -1 +1 @@
|
|
1 |
-
{"default": {"description": "ViGGO was designed for the task of data-to-text generation in chatbots (as opposed to task-oriented dialogue systems), with target responses being more conversational than information-seeking, yet constrained to the information presented in a meaning representation. The dataset, being relatively small and clean, can also serve for demonstrating transfer learning capabilities of neural models.\n", "citation": "@inproceedings{juraska-etal-2019-viggo,\n title = \"{V}i{GGO}: A Video Game Corpus for Data-To-Text Generation in Open-Domain Conversation\",\n author = \"Juraska, Juraj and\n Bowden, Kevin and\n Walker, Marilyn\",\n booktitle = \"Proceedings of the 12th International Conference on Natural Language Generation\",\n month = oct # \"{--}\" # nov,\n year = \"2019\",\n address = \"Tokyo, Japan\",\n publisher = \"Association for Computational Linguistics\",\n url = \"https://aclanthology.org/W19-8623\",\n doi = \"10.18653/v1/W19-8623\",\n pages = \"164--172\",\n}\n", "homepage": "https://nlds.soe.ucsc.edu/viggo", "license": "", "features": {"gem_id": {"dtype": "string", "id": null, "_type": "Value"}, "meaning_representation": {"dtype": "string", "id": null, "_type": "Value"}, "target": {"dtype": "string", "id": null, "_type": "Value"}, "references": [{"dtype": "string", "id": null, "_type": "Value"}]}, "post_processed": null, "supervised_keys": {"input": "meaning_representation", "output": "target"}, "task_templates": null, "builder_name": "viggo", "config_name": "default", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes":
|
|
|
1 |
+
{"default": {"description": "ViGGO was designed for the task of data-to-text generation in chatbots (as opposed to task-oriented dialogue systems), with target responses being more conversational than information-seeking, yet constrained to the information presented in a meaning representation. The dataset, being relatively small and clean, can also serve for demonstrating transfer learning capabilities of neural models.\n", "citation": "@inproceedings{juraska-etal-2019-viggo,\n title = \"{V}i{GGO}: A Video Game Corpus for Data-To-Text Generation in Open-Domain Conversation\",\n author = \"Juraska, Juraj and\n Bowden, Kevin and\n Walker, Marilyn\",\n booktitle = \"Proceedings of the 12th International Conference on Natural Language Generation\",\n month = oct # \"{--}\" # nov,\n year = \"2019\",\n address = \"Tokyo, Japan\",\n publisher = \"Association for Computational Linguistics\",\n url = \"https://aclanthology.org/W19-8623\",\n doi = \"10.18653/v1/W19-8623\",\n pages = \"164--172\",\n}\n", "homepage": "https://nlds.soe.ucsc.edu/viggo", "license": "", "features": {"gem_id": {"dtype": "string", "id": null, "_type": "Value"}, "meaning_representation": {"dtype": "string", "id": null, "_type": "Value"}, "target": {"dtype": "string", "id": null, "_type": "Value"}, "references": [{"dtype": "string", "id": null, "_type": "Value"}]}, "post_processed": null, "supervised_keys": {"input": "meaning_representation", "output": "target"}, "task_templates": null, "builder_name": "viggo", "config_name": "default", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 2048563, "num_examples": 5103, "dataset_name": "viggo"}, "validation": {"name": "validation", "num_bytes": 278970, "num_examples": 714, "dataset_name": "viggo"}, "test": {"name": "test", "num_bytes": 405327, "num_examples": 1083, "dataset_name": "viggo"}, "challenge_train_1_percent": {"name": "challenge_train_1_percent", "num_bytes": 19135, "num_examples": 50, "dataset_name": "viggo"}, "challenge_train_2_percent": {"name": "challenge_train_2_percent", "num_bytes": 42598, "num_examples": 103, "dataset_name": "viggo"}, "challenge_train_5_percent": {"name": "challenge_train_5_percent", "num_bytes": 109197, "num_examples": 256, "dataset_name": "viggo"}, "challenge_train_10_percent": {"name": "challenge_train_10_percent", "num_bytes": 210415, "num_examples": 510, "dataset_name": "viggo"}, "challenge_train_20_percent": {"name": "challenge_train_20_percent", "num_bytes": 431945, "num_examples": 1021, "dataset_name": "viggo"}}, "download_checksums": {"train.csv": {"num_bytes": 1370770, "checksum": "39d4bf7ef5b7b78c1c137d05d461e779884ad08cf53e0ee6cf63c3653c5f8aae"}, "validation.csv": {"num_bytes": 187448, "checksum": "c4d4b3b5075d84c1645cccd678e665e0d2a40226e3e9a383270a5c666006adbc"}, "test.csv": {"num_bytes": 268969, "checksum": "020f24199655b4c6f9246123263d2c6d3ea01189046150153dcc6cf8b914037a"}, "challenge_train_1_percent.csv": {"num_bytes": 11230, "checksum": "84a2540a71b5034c0d454899d6f68bfd2c0a48c7e3c6fb8e34723189bf818eaa"}, "challenge_train_2_percent.csv": {"num_bytes": 25113, "checksum": "17f3f49f21d31f616e570ed10441f3cfa7af4feef0789c0a182b1ac0c0ac47c1"}, "challenge_train_5_percent.csv": {"num_bytes": 64269, "checksum": "6c497a8bcb545ac55151c836a6cd5a3ae0d702d8394298d85154a41624d0e6dd"}, "challenge_train_10_percent.csv": {"num_bytes": 123238, "checksum": "1717d1481a285eb71cc4f0dff9766e6620419a7a206a81a97555b5c51c765e77"}, "challenge_train_20_percent.csv": {"num_bytes": 252901, "checksum": "7e02fe7c2be951d171142eb0568c433e5fb458c7332abf906d2e2ff2ac85049e"}}, "download_size": 2303938, "post_processing_size": null, "dataset_size": 3546150, "size_in_bytes": 5850088}}
|
viggo.py
CHANGED
@@ -75,7 +75,7 @@ class Viggo(datasets.GeneratorBasedBuilder):
|
|
75 |
reader = csv.DictReader(csvfile)
|
76 |
for id_, row in enumerate(reader):
|
77 |
yield id_, {
|
78 |
-
"gem_id": f"
|
79 |
"meaning_representation": row["mr"],
|
80 |
"target": row["ref"],
|
81 |
"references": [row["ref"]],
|
|
|
75 |
reader = csv.DictReader(csvfile)
|
76 |
for id_, row in enumerate(reader):
|
77 |
yield id_, {
|
78 |
+
"gem_id": f"viggo-{split}-{id_}",
|
79 |
"meaning_representation": row["mr"],
|
80 |
"target": row["ref"],
|
81 |
"references": [row["ref"]],
|