sileod commited on
Commit
5657d2f
1 Parent(s): fad468c

Create crowdflower.py

Browse files
Files changed (1) hide show
  1. crowdflower.py +210 -0
crowdflower.py ADDED
@@ -0,0 +1,210 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ # Lint as: python3
17
+ """Crowdflower datasets"""
18
+
19
+ from __future__ import absolute_import, division, print_function
20
+
21
+ import csv
22
+ import os
23
+ import textwrap
24
+
25
+ import six
26
+
27
+ import datasets
28
+
29
+
30
+ _crowdflower_CITATION = r"""
31
+ @inproceedings{van2012designing,
32
+ title={Designing a scalable crowdsourcing platform},
33
+ author={Van Pelt, Chris and Sorokin, Alex},
34
+ booktitle={Proceedings of the 2012 ACM SIGMOD International Conference on Management of Data},
35
+ pages={765--766},
36
+ year={2012}
37
+ }
38
+ """
39
+
40
+ _crowdflower_DESCRIPTION = """
41
+ Collection of crowdflower classification datasets
42
+ """
43
+
44
+ DATA_URL = "https://www.dropbox.com/s/ldrcdsv8d9qiwg0/crowdflower.zip?dl=1"
45
+
46
+ TASK_TO_LABELS = {'airline-sentiment': ['neutral', 'positive', 'negative'],
47
+ 'corporate-messaging': ['Information', 'Action', 'Exclude', 'Dialogue'],
48
+ 'economic-news': ['not sure', 'yes', 'no'],
49
+ 'political-media-audience': ['constituency', 'national'],
50
+ 'political-media-bias': ['partisan', 'neutral'],
51
+ 'political-media-message': ['information',
52
+ 'support',
53
+ 'policy',
54
+ 'constituency',
55
+ 'personal',
56
+ 'other',
57
+ 'media',
58
+ 'mobilization',
59
+ 'attack'],
60
+ 'sentiment_nuclear_power': ['Neutral / author is just sharing information',
61
+ 'Negative',
62
+ 'Tweet NOT related to nuclear energy',
63
+ 'Positive'],
64
+ 'text_emotion': ['sadness',
65
+ 'empty',
66
+ 'relief',
67
+ 'hate',
68
+ 'worry',
69
+ 'enthusiasm',
70
+ 'happiness',
71
+ 'neutral',
72
+ 'love',
73
+ 'fun',
74
+ 'anger',
75
+ 'surprise',
76
+ 'boredom'],
77
+ 'tweet_global_warming': ['Yes', 'No']}
78
+
79
+ def get_labels(task):
80
+ return TASK_TO_LABELS[task]
81
+
82
+ class crowdflowerConfig(datasets.BuilderConfig):
83
+ """BuilderConfig for crowdflower."""
84
+
85
+ def __init__(
86
+ self,
87
+ text_features,
88
+ label_classes=None,
89
+ process_label=lambda x: x,
90
+ **kwargs,
91
+ ):
92
+ """BuilderConfig for crowdflower.
93
+ Args:
94
+ text_features: `dict[string, string]`, map from the name of the feature
95
+ dict for each text field to the name of the column in the tsv file
96
+ label_column: `string`, name of the column in the tsv file corresponding
97
+ to the label
98
+ data_url: `string`, url to download the zip file from
99
+ data_dir: `string`, the path to the folder containing the tsv files in the
100
+ downloaded zip
101
+ citation: `string`, citation for the data set
102
+ url: `string`, url for information about the data set
103
+ label_classes: `list[string]`, the list of classes if the label is
104
+ categorical. If not provided, then the label will be of type
105
+ `datasets.Value('float32')`.
106
+ process_label: `Function[string, any]`, function taking in the raw value
107
+ of the label and processing it to the form required by the label feature
108
+ **kwargs: keyword arguments forwarded to super.
109
+ """
110
+
111
+ super(crowdflowerConfig, self).__init__(version=datasets.Version("1.0.0", ""), **kwargs)
112
+
113
+ self.text_features = text_features
114
+ self.label_column = "label"
115
+ self.label_classes = get_labels(self.name)
116
+ self.data_url = DATA_URL
117
+ self.data_dir = os.path.join("crowdflower", self.name)
118
+ self.citation = textwrap.dedent(_crowdflower_CITATION)
119
+ def process_label(x):
120
+ x=str(x)
121
+ if x=="Y":
122
+ return "Yes"
123
+ if x=="N":
124
+ return "No"
125
+ return x
126
+ self.process_label = process_label
127
+ self.description = ""
128
+ self.url = ""
129
+
130
+
131
+ class crowdflower(datasets.GeneratorBasedBuilder):
132
+
133
+ """The General Language Understanding Evaluation (crowdflower) benchmark."""
134
+
135
+ BUILDER_CONFIG_CLASS = crowdflowerConfig
136
+
137
+ BUILDER_CONFIGS = [
138
+ crowdflowerConfig(name="sentiment_nuclear_power",
139
+ text_features={"text": "text"},),
140
+ crowdflowerConfig(name="tweet_global_warming",
141
+ text_features={"text": "text"},),
142
+ crowdflowerConfig(name="airline-sentiment",
143
+ text_features={"text": "text"},),
144
+ crowdflowerConfig(name="corporate-messaging",
145
+ text_features={"text": "text"},),
146
+ crowdflowerConfig(name="economic-news",
147
+ text_features={"text": "text"},),
148
+ crowdflowerConfig(name="political-media-audience",
149
+ text_features={"text": "text"},),
150
+ crowdflowerConfig(name="political-media-bias",
151
+ text_features={"text": "text"},),
152
+ crowdflowerConfig(name="political-media-message",
153
+ text_features={"text": "text"},),
154
+ crowdflowerConfig(name="text_emotion",
155
+ text_features={"text": "text"},),
156
+ ]
157
+
158
+ def _info(self):
159
+ features = {text_feature: datasets.Value("string") for text_feature in six.iterkeys(self.config.text_features)}
160
+ if self.config.label_classes:
161
+ features["label"] = datasets.features.ClassLabel(names=self.config.label_classes)
162
+ else:
163
+ features["label"] = datasets.Value("float32")
164
+ features["idx"] = datasets.Value("int32")
165
+ return datasets.DatasetInfo(
166
+ description=_crowdflower_DESCRIPTION,
167
+ features=datasets.Features(features),
168
+ homepage=self.config.url,
169
+ citation=self.config.citation + "\n" + _crowdflower_CITATION,
170
+ )
171
+
172
+ def _split_generators(self, dl_manager):
173
+ dl_dir = dl_manager.download_and_extract(self.config.data_url)
174
+ data_dir = os.path.join(dl_dir, self.config.data_dir)
175
+
176
+ return [
177
+ datasets.SplitGenerator(
178
+ name=datasets.Split.TRAIN,
179
+ gen_kwargs={
180
+ "data_file": os.path.join(data_dir or "", "train.tsv"),
181
+ "split": "train",
182
+ },
183
+ ),
184
+ ]
185
+
186
+ def _generate_examples(self, data_file, split):
187
+
188
+ process_label = self.config.process_label
189
+ label_classes = self.config.label_classes
190
+
191
+ with open(data_file, encoding="latin-1") as f:
192
+ reader = csv.DictReader(f, delimiter="\t", quoting=csv.QUOTE_NONE)
193
+
194
+ for n, row in enumerate(reader):
195
+
196
+ example = {feat: row[col] for feat, col in six.iteritems(self.config.text_features)}
197
+ example["idx"] = n
198
+ #print(row)
199
+
200
+ if self.config.label_column in row:
201
+ label = row[self.config.label_column]
202
+ label = process_label(label)
203
+ if label_classes and label not in label_classes:
204
+ continue
205
+ example["label"] = label
206
+ else:
207
+ example["label"] = process_label(-1)
208
+ if not example["label"] or not example["text"]:
209
+ continue
210
+ yield example["idx"], example