Datasets:
Upload fair_rationales.py
Browse files- fair_rationales.py +7 -5
fair_rationales.py
CHANGED
@@ -58,6 +58,7 @@ class FairRationalesConfig(datasets.BuilderConfig):
|
|
58 |
|
59 |
def __init__(
|
60 |
self,
|
|
|
61 |
url,
|
62 |
data_url,
|
63 |
citation,
|
@@ -86,6 +87,7 @@ class FairRationalesConfig(datasets.BuilderConfig):
|
|
86 |
**kwargs: keyword arguments forwarded to super.
|
87 |
"""
|
88 |
super(FairRationalesConfig, self).__init__(version=datasets.Version("1.0.0", ""), **kwargs)
|
|
|
89 |
self.label_classes = label_classes
|
90 |
self.label_classes_original = label_classes_original
|
91 |
# self.attentioncheck = attentioncheck
|
@@ -103,7 +105,7 @@ class FairRationales(datasets.GeneratorBasedBuilder):
|
|
103 |
|
104 |
BUILDER_CONFIGS = [
|
105 |
FairRationalesConfig(
|
106 |
-
|
107 |
description=textwrap.dedent(
|
108 |
"""\
|
109 |
The Stanford Sentiment Treebank is a corpus with fully labeled parse trees that allows for a complete analysis of the compositional effects of sentiment in language.
|
@@ -150,7 +152,7 @@ class FairRationales(datasets.GeneratorBasedBuilder):
|
|
150 |
),
|
151 |
),
|
152 |
FairRationalesConfig(
|
153 |
-
|
154 |
description=textwrap.dedent(
|
155 |
"""\
|
156 |
DynaSent is an English-language benchmark task for ternary (positive/negative/neutral) sentiment analysis.
|
@@ -183,7 +185,7 @@ class FairRationales(datasets.GeneratorBasedBuilder):
|
|
183 |
),
|
184 |
),
|
185 |
FairRationalesConfig(
|
186 |
-
|
187 |
description=textwrap.dedent(
|
188 |
"""\
|
189 |
Common Sense Explanations (CoS-E) allows for training language models to automatically
|
@@ -236,7 +238,7 @@ class FairRationales(datasets.GeneratorBasedBuilder):
|
|
236 |
"originaldata_id": datasets.Value("string"),
|
237 |
"annotator_ID": datasets.Value("int64")
|
238 |
}
|
239 |
-
if self.config.
|
240 |
features["label"] = datasets.Value("string")
|
241 |
else:
|
242 |
features["label"] = datasets.ClassLabel(names=self.config.label_classes)
|
@@ -288,7 +290,7 @@ class FairRationales(datasets.GeneratorBasedBuilder):
|
|
288 |
}
|
289 |
for attribute_name, _ in self.config.attributes:
|
290 |
example[attribute_name] = data[attribute_name]
|
291 |
-
if self.config.
|
292 |
example["sst2_id"] = data["sst2_id"]
|
293 |
example["sst2_split"] = data["sst2_split"]
|
294 |
yield id_, example
|
|
|
58 |
|
59 |
def __init__(
|
60 |
self,
|
61 |
+
dataname,
|
62 |
url,
|
63 |
data_url,
|
64 |
citation,
|
|
|
87 |
**kwargs: keyword arguments forwarded to super.
|
88 |
"""
|
89 |
super(FairRationalesConfig, self).__init__(version=datasets.Version("1.0.0", ""), **kwargs)
|
90 |
+
self.dataname = dataname
|
91 |
self.label_classes = label_classes
|
92 |
self.label_classes_original = label_classes_original
|
93 |
# self.attentioncheck = attentioncheck
|
|
|
105 |
|
106 |
BUILDER_CONFIGS = [
|
107 |
FairRationalesConfig(
|
108 |
+
dataname="sst2",
|
109 |
description=textwrap.dedent(
|
110 |
"""\
|
111 |
The Stanford Sentiment Treebank is a corpus with fully labeled parse trees that allows for a complete analysis of the compositional effects of sentiment in language.
|
|
|
152 |
),
|
153 |
),
|
154 |
FairRationalesConfig(
|
155 |
+
dataname="dynasent",
|
156 |
description=textwrap.dedent(
|
157 |
"""\
|
158 |
DynaSent is an English-language benchmark task for ternary (positive/negative/neutral) sentiment analysis.
|
|
|
185 |
),
|
186 |
),
|
187 |
FairRationalesConfig(
|
188 |
+
dataname="cose",
|
189 |
description=textwrap.dedent(
|
190 |
"""\
|
191 |
Common Sense Explanations (CoS-E) allows for training language models to automatically
|
|
|
238 |
"originaldata_id": datasets.Value("string"),
|
239 |
"annotator_ID": datasets.Value("int64")
|
240 |
}
|
241 |
+
if self.config.dataname == "cose":
|
242 |
features["label"] = datasets.Value("string")
|
243 |
else:
|
244 |
features["label"] = datasets.ClassLabel(names=self.config.label_classes)
|
|
|
290 |
}
|
291 |
for attribute_name, _ in self.config.attributes:
|
292 |
example[attribute_name] = data[attribute_name]
|
293 |
+
if self.config.dataname == "sst2":
|
294 |
example["sst2_id"] = data["sst2_id"]
|
295 |
example["sst2_split"] = data["sst2_split"]
|
296 |
yield id_, example
|