Datasets:

Languages:
English
Multilinguality:
monolingual
Size Categories:
100K<n<1M
Language Creators:
found
Annotations Creators:
machine-generated
ArXiv:
Tags:
License:
albertvillanova HF staff commited on
Commit
1e853eb
1 Parent(s): 00b0e4a

Convert dataset to Parquet (#5)

Browse files

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

README.md CHANGED
@@ -1,15 +1,14 @@
1
  ---
2
  annotations_creators:
3
  - machine-generated
4
- language:
5
- - en
6
  language_creators:
7
  - found
 
 
8
  license:
9
  - unknown
10
  multilinguality:
11
  - monolingual
12
- pretty_name: CompGuessWhat?!
13
  size_categories:
14
  - 100K<n<1M
15
  source_datasets:
@@ -19,6 +18,7 @@ task_categories:
19
  task_ids:
20
  - visual-question-answering
21
  paperswithcode_id: compguesswhat
 
22
  dataset_info:
23
  - config_name: compguesswhat-original
24
  features:
@@ -84,16 +84,16 @@ dataset_info:
84
  sequence: float32
85
  splits:
86
  - name: train
87
- num_bytes: 126548689
88
  num_examples: 46341
89
  - name: validation
90
- num_bytes: 26055261
91
  num_examples: 9738
92
  - name: test
93
- num_bytes: 25981593
94
  num_examples: 9621
95
- download_size: 107201655
96
- dataset_size: 178585543
97
  - config_name: compguesswhat-zero_shot
98
  features:
99
  - name: id
@@ -159,19 +159,38 @@ dataset_info:
159
  dtype: string
160
  splits:
161
  - name: nd_valid
162
- num_bytes: 13557059
163
  num_examples: 5343
164
  - name: nd_test
165
- num_bytes: 36352201
166
  num_examples: 13836
167
  - name: od_valid
168
- num_bytes: 14093233
169
  num_examples: 5372
170
  - name: od_test
171
- num_bytes: 33049755
172
  num_examples: 13300
173
- download_size: 4845966
174
- dataset_size: 97052248
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
175
  ---
176
 
177
  # Dataset Card for "compguesswhat"
1
  ---
2
  annotations_creators:
3
  - machine-generated
 
 
4
  language_creators:
5
  - found
6
+ language:
7
+ - en
8
  license:
9
  - unknown
10
  multilinguality:
11
  - monolingual
 
12
  size_categories:
13
  - 100K<n<1M
14
  source_datasets:
18
  task_ids:
19
  - visual-question-answering
20
  paperswithcode_id: compguesswhat
21
+ pretty_name: CompGuessWhat?!
22
  dataset_info:
23
  - config_name: compguesswhat-original
24
  features:
84
  sequence: float32
85
  splits:
86
  - name: train
87
+ num_bytes: 123556580
88
  num_examples: 46341
89
  - name: validation
90
+ num_bytes: 25441428
91
  num_examples: 9738
92
  - name: test
93
+ num_bytes: 25369227
94
  num_examples: 9621
95
+ download_size: 105349759
96
+ dataset_size: 174367235
97
  - config_name: compguesswhat-zero_shot
98
  features:
99
  - name: id
159
  dtype: string
160
  splits:
161
  - name: nd_valid
162
+ num_bytes: 13510589
163
  num_examples: 5343
164
  - name: nd_test
165
+ num_bytes: 36228021
166
  num_examples: 13836
167
  - name: od_valid
168
+ num_bytes: 14051972
169
  num_examples: 5372
170
  - name: od_test
171
+ num_bytes: 32950869
172
  num_examples: 13300
173
+ download_size: 6548812
174
+ dataset_size: 96741451
175
+ configs:
176
+ - config_name: compguesswhat-original
177
+ data_files:
178
+ - split: train
179
+ path: compguesswhat-original/train-*
180
+ - split: validation
181
+ path: compguesswhat-original/validation-*
182
+ - split: test
183
+ path: compguesswhat-original/test-*
184
+ - config_name: compguesswhat-zero_shot
185
+ data_files:
186
+ - split: nd_valid
187
+ path: compguesswhat-zero_shot/nd_valid-*
188
+ - split: nd_test
189
+ path: compguesswhat-zero_shot/nd_test-*
190
+ - split: od_valid
191
+ path: compguesswhat-zero_shot/od_valid-*
192
+ - split: od_test
193
+ path: compguesswhat-zero_shot/od_test-*
194
  ---
195
 
196
  # Dataset Card for "compguesswhat"
compguesswhat-original/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:16b2a80dd58e66aad5fe8857c95a812e2206ae05a55ec6cfa89d38518577415a
3
+ size 15268653
compguesswhat-original/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a4099d3ab1826c0acbf47710ffa8197cf52204cfaa6df44149c1d60ee01dd41f
3
+ size 74763809
compguesswhat-original/validation-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:024d592321b0bc7f7f8d6419b1f45f8e188875f4bf187a59037c694b6e806784
3
+ size 15317297
compguesswhat-zero_shot/nd_test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c7bfb006600e91cf151b644b1495c5688268997cd024c9ce832d7500302d690e
3
+ size 2434268
compguesswhat-zero_shot/nd_valid-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:03d7d02f63bac35c043d2e73f1cceec10d365440f165348ff86dc8cf3fc9fcd4
3
+ size 929562
compguesswhat-zero_shot/od_test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:67eee5a8432934369fa4d4eb33e7fd11e4b57fb85bd2d5e826e46905b7d1a54b
3
+ size 2255581
compguesswhat-zero_shot/od_valid-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e6b4ef094118fb380e70a83dbcc2fa1c43cc56f655f48d419b8bfbd72bf795ac
3
+ size 929401
compguesswhat.py DELETED
@@ -1,254 +0,0 @@
1
- import gzip
2
- import json
3
- import os
4
-
5
- import datasets
6
-
7
-
8
- class CompguesswhatConfig(datasets.BuilderConfig):
9
- """BuilderConfig for CompGuessWhat?!"""
10
-
11
- def __init__(self, data_url, splits, gameplay_scenario, **kwargs):
12
- """
13
-
14
- Args:
15
- gameplay_scenario: to specify if we want to load original CompGuessWhat?! split ('original') or
16
- the zero-shot reference games based on NOCAPS images ('zero_shot')
17
- **kwargs: keyword arguments forwarded to super.
18
- """
19
- super(CompguesswhatConfig, self).__init__(
20
- version=datasets.Version("0.2.0", "Second CompGuessWhat?! release"), **kwargs
21
- )
22
- assert gameplay_scenario in (
23
- "original",
24
- "zero_shot",
25
- ), "Invalid choice for parameter 'gameplay_scenario': {gameplay_scenario}. Valid values are ('original', 'zero_shot')."
26
-
27
- self.gameplay_scenario = gameplay_scenario
28
- self.splits = splits
29
- self.data_url = data_url
30
-
31
-
32
- class Compguesswhat(datasets.GeneratorBasedBuilder):
33
- _CITATION = """\
34
- @inproceedings{suglia2020compguesswhat,
35
- title={CompGuessWhat?!: a Multi-task Evaluation Framework for Grounded Language Learning},
36
- author={Suglia, Alessandro, Konstas, Ioannis, Vanzo, Andrea, Bastianelli, Emanuele, Desmond Elliott, Stella Frank and Oliver Lemon},
37
- booktitle={Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics},
38
- year={2020}
39
- }
40
- """
41
-
42
- _DESCRIPTION = """
43
- CompGuessWhat?! is an instance of a multi-task framework for evaluating the quality of learned neural representations,
44
- in particular concerning attribute grounding. Use this dataset if you want to use the set of games whose reference
45
- scene is an image in VisualGenome. Visit the website for more details: https://compguesswhat.github.io
46
- """
47
-
48
- BUILDER_CONFIGS = [
49
- CompguesswhatConfig(
50
- name="compguesswhat-original",
51
- gameplay_scenario="original",
52
- description="CompGuessWhat?! subset of games from the original GuessWhat?! dataset",
53
- data_url="https://www.dropbox.com/s/qd9wlydpkpmq8rr/compguesswhat-original.zip?dl=1",
54
- splits={
55
- "train": "compguesswhat.train.jsonl.gz",
56
- "valid": "compguesswhat.valid.jsonl.gz",
57
- "test": "compguesswhat.test.jsonl.gz",
58
- },
59
- ),
60
- CompguesswhatConfig(
61
- name="compguesswhat-zero_shot",
62
- gameplay_scenario="zero_shot",
63
- description="CompGuessWhat?! reference set of games for zero-shot evaluation using NOCAPS images",
64
- data_url="https://www.dropbox.com/s/f5o2t7aiok7kpcm/compguesswhat-zero_shot.zip?dl=1",
65
- splits={
66
- "nd_valid": "compguesswhat.nd_valid.jsonl.gz",
67
- "nd_test": "compguesswhat.nd_test.jsonl.gz",
68
- "od_valid": "compguesswhat.od_valid.jsonl.gz",
69
- "od_test": "compguesswhat.od_test.jsonl.gz",
70
- },
71
- ),
72
- ]
73
-
74
- VERSION = datasets.Version("0.2.0")
75
-
76
- def _info(self):
77
- if self.config.gameplay_scenario == "original":
78
- return datasets.DatasetInfo(
79
- # This is the description that will appear on the datasets page.
80
- description=self._DESCRIPTION,
81
- # datasets.features.FeatureConnectors
82
- features=datasets.Features(
83
- {
84
- "id": datasets.Value("int32"),
85
- "target_id": datasets.Value("int32"),
86
- "timestamp": datasets.Value("string"),
87
- "status": datasets.Value("string"),
88
- "image": {
89
- # this is the image ID in GuessWhat?! which corresponds to the MSCOCO id
90
- "id": datasets.Value("int32"),
91
- "file_name": datasets.Value("string"),
92
- "flickr_url": datasets.Value("string"),
93
- "coco_url": datasets.Value("string"),
94
- "height": datasets.Value("int32"),
95
- "width": datasets.Value("int32"),
96
- # this field represents the corresponding image metadata that can be found in VisualGenome
97
- # in the file image_data.json
98
- # We copy it over so that we avoid any confusion or possible wrong URL
99
- # Please use the original image files to resolve photos
100
- "visual_genome": {
101
- "width": datasets.Value("int32"),
102
- "height": datasets.Value("int32"),
103
- "url": datasets.Value("string"),
104
- "coco_id": datasets.Value("int32"),
105
- # this is the actual VisualGenome image ID
106
- # because we can't rely store it as an integer we same it as string
107
- "flickr_id": datasets.Value("string"),
108
- "image_id": datasets.Value("string"),
109
- },
110
- },
111
- "qas": datasets.features.Sequence(
112
- {
113
- "question": datasets.Value("string"),
114
- "answer": datasets.Value("string"),
115
- "id": datasets.Value("int32"),
116
- }
117
- ),
118
- "objects": datasets.features.Sequence(
119
- {
120
- "id": datasets.Value("int32"),
121
- "bbox": datasets.Sequence(datasets.Value("float32"), length=4),
122
- "category": datasets.Value("string"),
123
- "area": datasets.Value("float32"),
124
- "category_id": datasets.Value("int32"),
125
- "segment": datasets.features.Sequence(
126
- datasets.features.Sequence(datasets.Value("float32"))
127
- ),
128
- }
129
- ),
130
- }
131
- ),
132
- # If there's a common (input, target) tuple from the features,
133
- # specify them here. They'll be used if as_supervised=True in
134
- # builder.as_dataset.
135
- supervised_keys=None,
136
- # Homepage of the dataset for documentation
137
- homepage="https://compguesswhat.github.io/",
138
- citation=self._CITATION,
139
- )
140
- elif self.config.gameplay_scenario == "zero_shot":
141
- return datasets.DatasetInfo(
142
- # This is the description that will appear on the datasets page.
143
- description=self._DESCRIPTION,
144
- # datasets.features.FeatureConnectors
145
- features=datasets.Features(
146
- {
147
- "id": datasets.Value("int32"),
148
- "target_id": datasets.Value("string"),
149
- "status": datasets.Value("string"),
150
- "image": {
151
- "id": datasets.Value("int32"),
152
- "file_name": datasets.Value("string"),
153
- "coco_url": datasets.Value("string"),
154
- "height": datasets.Value("int32"),
155
- "width": datasets.Value("int32"),
156
- "license": datasets.Value("int32"),
157
- "open_images_id": datasets.Value("string"),
158
- "date_captured": datasets.Value("string"),
159
- },
160
- "objects": datasets.features.Sequence(
161
- {
162
- "id": datasets.Value("string"),
163
- "bbox": datasets.Sequence(datasets.Value("float32"), length=4),
164
- "category": datasets.Value("string"),
165
- "area": datasets.Value("float32"),
166
- "category_id": datasets.Value("int32"),
167
- "IsOccluded": datasets.Value("int32"),
168
- "IsTruncated": datasets.Value("int32"),
169
- "segment": datasets.features.Sequence(
170
- {
171
- "MaskPath": datasets.Value("string"),
172
- "LabelName": datasets.Value("string"),
173
- "BoxID": datasets.Value("string"),
174
- "BoxXMin": datasets.Value("string"),
175
- "BoxXMax": datasets.Value("string"),
176
- "BoxYMin": datasets.Value("string"),
177
- "BoxYMax": datasets.Value("string"),
178
- "PredictedIoU": datasets.Value("string"),
179
- "Clicks": datasets.Value("string"),
180
- }
181
- ),
182
- }
183
- ),
184
- }
185
- ),
186
- # If there's a common (input, target) tuple from the features,
187
- # specify them here. They'll be used if as_supervised=True in
188
- # builder.as_dataset.
189
- supervised_keys=None,
190
- # Homepage of the dataset for documentation
191
- homepage="https://compguesswhat.github.io/",
192
- citation=self._CITATION,
193
- )
194
-
195
- def _split_generators(self, dl_manager):
196
- """Returns SplitGenerators."""
197
- dl_dir = dl_manager.download_and_extract(self.config.data_url)
198
-
199
- splits_gen = []
200
-
201
- for split_id, split_filename in self.config.splits.items():
202
- if self.config.gameplay_scenario == "original":
203
- if "train" in split_id:
204
- split_name = datasets.Split.TRAIN
205
- elif "valid" in split_id:
206
- split_name = datasets.Split.VALIDATION
207
- elif "test" in split_id:
208
- split_name = datasets.Split.TEST
209
- else:
210
- split_name = datasets.Split(split_id)
211
-
212
- full_split_name = "-".join(["compguesswhat", self.config.gameplay_scenario])
213
- splits_gen.append(
214
- datasets.SplitGenerator(
215
- name=split_name,
216
- gen_kwargs={
217
- "filepath": os.path.join(
218
- dl_dir,
219
- full_split_name,
220
- self.VERSION.version_str,
221
- split_filename,
222
- )
223
- },
224
- )
225
- )
226
-
227
- return splits_gen
228
-
229
- def _generate_examples(self, filepath):
230
- def _extract_game_tuple(data):
231
- data = data.decode("utf-8")
232
- game = json.loads(data.strip("\n"))
233
-
234
- # we refactor the data structure a bit to fit with the new version
235
- game["target_id"] = game["object_id"]
236
- if "object_id" in game:
237
- del game["object_id"]
238
-
239
- if "questioner_id" in game:
240
- del game["questioner_id"]
241
- ###
242
-
243
- if "visual_genome" in game["image"]:
244
- # We need to cast it to string so that we avoid issues with int size
245
- game["image"]["visual_genome"]["image_id"] = str(game["image"]["visual_genome"]["image_id"])
246
- game["image"]["visual_genome"]["flickr_id"] = str(game["image"]["visual_genome"]["flickr_id"])
247
-
248
- return game["id"], game
249
-
250
- """Yields examples."""
251
- with open(filepath, "rb") as gzip_file:
252
- with gzip.open(gzip_file) as in_file:
253
- for data in in_file:
254
- yield _extract_game_tuple(data)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
create_dummy_data.py DELETED
@@ -1,96 +0,0 @@
1
- import gzip
2
- import json
3
- import logging
4
- import os
5
- from argparse import ArgumentParser
6
-
7
-
8
- parser = ArgumentParser()
9
- parser.add_argument(
10
- "-d",
11
- "--data_path",
12
- type=str,
13
- required=True,
14
- help="Data path containing the CompGuessWhat?! datasets (files with 'jsonl.gz' extension)",
15
- )
16
- parser.add_argument(
17
- "--examples",
18
- type=int,
19
- default=5,
20
- help="Number of games to consider in the dummy dataset",
21
- )
22
- original_data_files = {
23
- "train": "compguesswhat.train.jsonl.gz",
24
- "valid": "compguesswhat.valid.jsonl.gz",
25
- "test": "compguesswhat.test.jsonl.gz",
26
- }
27
-
28
- zs_data_files = {
29
- "nd_valid": "compguesswhat.nd_valid.jsonl.gz",
30
- "od_valid": "compguesswhat.od_valid.jsonl.gz",
31
- "nd_test": "compguesswhat.nd_test.jsonl.gz",
32
- "od_test": "compguesswhat.od_test.jsonl.gz",
33
- }
34
-
35
- COMPGUESSWHAT_ROOT = "datasets/compguesswhat/"
36
-
37
-
38
- def create_dummy_data_for_split(data_path, dataset_name, dataset_version, data_files):
39
- full_dataset_name = "-".join(["compguesswhat", dataset_name])
40
- dummy_data_path = os.path.join(
41
- COMPGUESSWHAT_ROOT,
42
- "dummy",
43
- full_dataset_name,
44
- dataset_version,
45
- "dummy_data",
46
- full_dataset_name,
47
- dataset_version,
48
- )
49
-
50
- if not os.path.exists(dummy_data_path):
51
- os.makedirs(dummy_data_path)
52
-
53
- for split_name, split_file in data_files.items():
54
- logging.info(f"Generating dummy data for split {split_name} (num. examples = {args.examples})")
55
-
56
- split_filepath = os.path.join(data_path, full_dataset_name, dataset_version, split_file)
57
- logging.info(f"Reading split file {split_filepath}")
58
- with gzip.open(split_filepath) as in_file:
59
- dummy_filepath = os.path.join(dummy_data_path, split_file)
60
- with gzip.open(dummy_filepath, mode="w") as out_file:
61
- for i, line in enumerate(in_file):
62
- if i > args.examples:
63
- break
64
-
65
- data = json.loads(line.strip())
66
- out_file.write(json.dumps(data).encode("utf-8"))
67
- out_file.write(b"\n")
68
-
69
-
70
- def main(args):
71
- # args.data_path is the directory containing the already downloaded dataset files
72
- # we assume that the dataset test was successful and we have the file dataset_info.json
73
- dataset_info_path = os.path.join(COMPGUESSWHAT_ROOT, "dataset_infos.json")
74
-
75
- if not os.path.exists(dataset_info_path):
76
- raise ValueError(
77
- "The file 'dataset_info.json' doesn't exists. Make sure that you run the dataset tests via datasets-cli."
78
- )
79
-
80
- with open(dataset_info_path, encoding="utf-8") as in_file:
81
- dataset_info = json.load(in_file)
82
-
83
- dataset_version = dataset_info["compguesswhat-original"]["version"]["version_str"]
84
-
85
- logging.info(f"Creating dummy data for CompGuessWhat?! {dataset_version}")
86
-
87
- logging.info("Original dataset...")
88
- create_dummy_data_for_split(args.data_path, "original", dataset_version, original_data_files)
89
-
90
- logging.info("Zero-shot dataset...")
91
- create_dummy_data_for_split(args.data_path, "zero_shot", dataset_version, zs_data_files)
92
-
93
-
94
- if __name__ == "__main__":
95
- args = parser.parse_args()
96
- main(args)