refactor JGLUE.py (#5)
Browse files* refactor JGLUE.py
* fix for the CI
JGLUE.py
CHANGED
@@ -6,6 +6,7 @@ from typing import Dict, List, Optional, Union
|
|
6 |
|
7 |
import datasets as ds
|
8 |
import pandas as pd
|
|
|
9 |
|
10 |
_CITATION = """\
|
11 |
@inproceedings{kurihara-etal-2022-jglue,
|
@@ -80,7 +81,7 @@ _URLS = {
|
|
80 |
}
|
81 |
|
82 |
|
83 |
-
def
|
84 |
features = ds.Features(
|
85 |
{
|
86 |
"sentence_pair_id": ds.Value("string"),
|
@@ -90,10 +91,16 @@ def features_jsts() -> ds.Features:
|
|
90 |
"label": ds.Value("float"),
|
91 |
}
|
92 |
)
|
93 |
-
return
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
|
95 |
|
96 |
-
def
|
97 |
features = ds.Features(
|
98 |
{
|
99 |
"sentence_pair_id": ds.Value("string"),
|
@@ -105,10 +112,17 @@ def features_jnli() -> ds.Features:
|
|
105 |
),
|
106 |
}
|
107 |
)
|
108 |
-
return
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
109 |
|
110 |
|
111 |
-
def
|
112 |
features = ds.Features(
|
113 |
{
|
114 |
"id": ds.Value("string"),
|
@@ -121,10 +135,24 @@ def features_jsquad() -> ds.Features:
|
|
121 |
"is_impossible": ds.Value("bool"),
|
122 |
}
|
123 |
)
|
124 |
-
return
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
125 |
|
126 |
|
127 |
-
def
|
128 |
features = ds.Features(
|
129 |
{
|
130 |
"q_id": ds.Value("int64"),
|
@@ -134,13 +162,22 @@ def features_jcommonsenseqa() -> ds.Features:
|
|
134 |
"choice2": ds.Value("string"),
|
135 |
"choice3": ds.Value("string"),
|
136 |
"choice4": ds.Value("string"),
|
137 |
-
"label": ds.
|
|
|
|
|
|
|
138 |
}
|
139 |
)
|
140 |
-
return
|
|
|
|
|
|
|
|
|
|
|
|
|
141 |
|
142 |
|
143 |
-
def
|
144 |
features = ds.Features(
|
145 |
{
|
146 |
"sentence": ds.Value("string"),
|
@@ -150,7 +187,13 @@ def features_marc_ja() -> ds.Features:
|
|
150 |
"review_id": ds.Value("string"),
|
151 |
}
|
152 |
)
|
153 |
-
return
|
|
|
|
|
|
|
|
|
|
|
|
|
154 |
|
155 |
|
156 |
class MarcJaConfig(ds.BuilderConfig):
|
@@ -439,60 +482,118 @@ class JGLUE(ds.GeneratorBasedBuilder):
|
|
439 |
|
440 |
def _info(self) -> ds.DatasetInfo:
|
441 |
if self.config.name == "JSTS":
|
442 |
-
|
443 |
elif self.config.name == "JNLI":
|
444 |
-
|
445 |
elif self.config.name == "JSQuAD":
|
446 |
-
|
447 |
elif self.config.name == "JCommonsenseQA":
|
448 |
-
|
449 |
elif self.config.name == "MARC-ja":
|
450 |
-
|
451 |
else:
|
452 |
raise ValueError(f"Invalid config name: {self.config.name}")
|
453 |
|
454 |
-
|
455 |
-
|
456 |
-
|
457 |
-
|
458 |
-
|
459 |
-
|
|
|
|
|
|
|
|
|
|
|
460 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
461 |
|
462 |
-
def
|
463 |
file_paths = dl_manager.download_and_extract(_URLS[self.config.name])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
464 |
|
|
|
465 |
if self.config.name == "MARC-ja":
|
466 |
-
|
467 |
-
label_conv_review_id_list = file_paths["label_conv_review_id_list"]
|
468 |
-
|
469 |
-
split_dfs = preprocess_for_marc_ja(
|
470 |
-
config=self.config,
|
471 |
-
data_file_path=file_paths["data"],
|
472 |
-
filter_review_id_list_paths=filter_review_id_list,
|
473 |
-
label_conv_review_id_list_paths=label_conv_review_id_list,
|
474 |
-
)
|
475 |
-
return [
|
476 |
-
ds.SplitGenerator(
|
477 |
-
name=ds.Split.TRAIN,
|
478 |
-
gen_kwargs={"split_df": split_dfs["train"]},
|
479 |
-
),
|
480 |
-
ds.SplitGenerator(
|
481 |
-
name=ds.Split.VALIDATION,
|
482 |
-
gen_kwargs={"split_df": split_dfs["valid"]},
|
483 |
-
),
|
484 |
-
]
|
485 |
else:
|
486 |
-
return
|
487 |
-
|
488 |
-
|
489 |
-
|
490 |
-
|
491 |
-
|
492 |
-
|
493 |
-
|
494 |
-
|
495 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
496 |
|
497 |
def _generate_examples(
|
498 |
self,
|
@@ -500,46 +601,13 @@ class JGLUE(ds.GeneratorBasedBuilder):
|
|
500 |
split_df: Optional[pd.DataFrame] = None,
|
501 |
):
|
502 |
if self.config.name == "MARC-ja":
|
503 |
-
|
504 |
-
raise ValueError(f"Invalid preprocessing for {self.config.name}")
|
505 |
|
506 |
-
|
507 |
-
|
508 |
-
|
|
|
|
|
509 |
|
510 |
else:
|
511 |
-
|
512 |
-
raise ValueError(f"Invalid argument for {self.config.name}")
|
513 |
-
|
514 |
-
if self.config.name == "JSQuAD":
|
515 |
-
with open(file_path, "r") as rf:
|
516 |
-
json_data = json.load(rf)
|
517 |
-
|
518 |
-
for json_dict in json_data["data"]:
|
519 |
-
title = json_dict["title"]
|
520 |
-
paragraphs = json_dict["paragraphs"]
|
521 |
-
for paragraph in paragraphs:
|
522 |
-
context = paragraph["context"]
|
523 |
-
questions = paragraph["qas"]
|
524 |
-
for question_dict in questions:
|
525 |
-
q_id = question_dict["id"]
|
526 |
-
question = question_dict["question"]
|
527 |
-
answers = question_dict["answers"]
|
528 |
-
is_impossible = question_dict["is_impossible"]
|
529 |
-
|
530 |
-
example_dict = {
|
531 |
-
"id": q_id,
|
532 |
-
"title": title,
|
533 |
-
"context": context,
|
534 |
-
"question": question,
|
535 |
-
"answers": answers,
|
536 |
-
"is_impossible": is_impossible,
|
537 |
-
}
|
538 |
-
|
539 |
-
yield q_id, example_dict
|
540 |
-
|
541 |
-
else:
|
542 |
-
with open(file_path, "r") as rf:
|
543 |
-
for i, line in enumerate(rf):
|
544 |
-
json_dict = json.loads(line)
|
545 |
-
yield i, json_dict
|
|
|
6 |
|
7 |
import datasets as ds
|
8 |
import pandas as pd
|
9 |
+
from datasets.tasks import QuestionAnsweringExtractive
|
10 |
|
11 |
_CITATION = """\
|
12 |
@inproceedings{kurihara-etal-2022-jglue,
|
|
|
81 |
}
|
82 |
|
83 |
|
84 |
+
def dataset_info_jsts() -> ds.Features:
|
85 |
features = ds.Features(
|
86 |
{
|
87 |
"sentence_pair_id": ds.Value("string"),
|
|
|
91 |
"label": ds.Value("float"),
|
92 |
}
|
93 |
)
|
94 |
+
return ds.DatasetInfo(
|
95 |
+
description=_DESCRIPTION,
|
96 |
+
citation=_CITATION,
|
97 |
+
homepage=_HOMEPAGE,
|
98 |
+
license=_LICENSE,
|
99 |
+
features=features,
|
100 |
+
)
|
101 |
|
102 |
|
103 |
+
def dataset_info_jnli() -> ds.Features:
|
104 |
features = ds.Features(
|
105 |
{
|
106 |
"sentence_pair_id": ds.Value("string"),
|
|
|
112 |
),
|
113 |
}
|
114 |
)
|
115 |
+
return ds.DatasetInfo(
|
116 |
+
description=_DESCRIPTION,
|
117 |
+
citation=_CITATION,
|
118 |
+
homepage=_HOMEPAGE,
|
119 |
+
license=_LICENSE,
|
120 |
+
features=features,
|
121 |
+
supervised_keys=None,
|
122 |
+
)
|
123 |
|
124 |
|
125 |
+
def dataset_info_jsquad() -> ds.Features:
|
126 |
features = ds.Features(
|
127 |
{
|
128 |
"id": ds.Value("string"),
|
|
|
135 |
"is_impossible": ds.Value("bool"),
|
136 |
}
|
137 |
)
|
138 |
+
return ds.DatasetInfo(
|
139 |
+
description=_DESCRIPTION,
|
140 |
+
citation=_CITATION,
|
141 |
+
homepage=_HOMEPAGE,
|
142 |
+
license=_LICENSE,
|
143 |
+
features=features,
|
144 |
+
supervised_keys=None,
|
145 |
+
task_templates=[
|
146 |
+
QuestionAnsweringExtractive(
|
147 |
+
question_column="question",
|
148 |
+
context_column="context",
|
149 |
+
answers_column="answers",
|
150 |
+
)
|
151 |
+
],
|
152 |
+
)
|
153 |
|
154 |
|
155 |
+
def dataset_info_jcommonsenseqa() -> ds.Features:
|
156 |
features = ds.Features(
|
157 |
{
|
158 |
"q_id": ds.Value("int64"),
|
|
|
162 |
"choice2": ds.Value("string"),
|
163 |
"choice3": ds.Value("string"),
|
164 |
"choice4": ds.Value("string"),
|
165 |
+
"label": ds.ClassLabel(
|
166 |
+
num_classes=5,
|
167 |
+
names=["choice0", "choice1", "choice2", "choice3", "choice4"],
|
168 |
+
),
|
169 |
}
|
170 |
)
|
171 |
+
return ds.DatasetInfo(
|
172 |
+
description=_DESCRIPTION,
|
173 |
+
citation=_CITATION,
|
174 |
+
homepage=_HOMEPAGE,
|
175 |
+
license=_LICENSE,
|
176 |
+
features=features,
|
177 |
+
)
|
178 |
|
179 |
|
180 |
+
def dataset_info_marc_ja() -> ds.Features:
|
181 |
features = ds.Features(
|
182 |
{
|
183 |
"sentence": ds.Value("string"),
|
|
|
187 |
"review_id": ds.Value("string"),
|
188 |
}
|
189 |
)
|
190 |
+
return ds.DatasetInfo(
|
191 |
+
description=_DESCRIPTION,
|
192 |
+
citation=_CITATION,
|
193 |
+
homepage=_HOMEPAGE,
|
194 |
+
license=_LICENSE,
|
195 |
+
features=features,
|
196 |
+
)
|
197 |
|
198 |
|
199 |
class MarcJaConfig(ds.BuilderConfig):
|
|
|
482 |
|
483 |
def _info(self) -> ds.DatasetInfo:
|
484 |
if self.config.name == "JSTS":
|
485 |
+
return dataset_info_jsts()
|
486 |
elif self.config.name == "JNLI":
|
487 |
+
return dataset_info_jnli()
|
488 |
elif self.config.name == "JSQuAD":
|
489 |
+
return dataset_info_jsquad()
|
490 |
elif self.config.name == "JCommonsenseQA":
|
491 |
+
return dataset_info_jcommonsenseqa()
|
492 |
elif self.config.name == "MARC-ja":
|
493 |
+
return dataset_info_marc_ja()
|
494 |
else:
|
495 |
raise ValueError(f"Invalid config name: {self.config.name}")
|
496 |
|
497 |
+
def __split_generators_marc_ja(self, dl_manager: ds.DownloadManager):
|
498 |
+
file_paths = dl_manager.download_and_extract(_URLS[self.config.name])
|
499 |
+
|
500 |
+
filter_review_id_list = file_paths["filter_review_id_list"]
|
501 |
+
label_conv_review_id_list = file_paths["label_conv_review_id_list"]
|
502 |
+
|
503 |
+
split_dfs = preprocess_for_marc_ja(
|
504 |
+
config=self.config,
|
505 |
+
data_file_path=file_paths["data"],
|
506 |
+
filter_review_id_list_paths=filter_review_id_list,
|
507 |
+
label_conv_review_id_list_paths=label_conv_review_id_list,
|
508 |
)
|
509 |
+
return [
|
510 |
+
ds.SplitGenerator(
|
511 |
+
name=ds.Split.TRAIN,
|
512 |
+
gen_kwargs={"split_df": split_dfs["train"]},
|
513 |
+
),
|
514 |
+
ds.SplitGenerator(
|
515 |
+
name=ds.Split.VALIDATION,
|
516 |
+
gen_kwargs={"split_df": split_dfs["valid"]},
|
517 |
+
),
|
518 |
+
]
|
519 |
|
520 |
+
def __split_generators(self, dl_manager: ds.DownloadManager):
|
521 |
file_paths = dl_manager.download_and_extract(_URLS[self.config.name])
|
522 |
+
return [
|
523 |
+
ds.SplitGenerator(
|
524 |
+
name=ds.Split.TRAIN,
|
525 |
+
gen_kwargs={"file_path": file_paths["train"]},
|
526 |
+
),
|
527 |
+
ds.SplitGenerator(
|
528 |
+
name=ds.Split.VALIDATION,
|
529 |
+
gen_kwargs={"file_path": file_paths["valid"]},
|
530 |
+
),
|
531 |
+
]
|
532 |
|
533 |
+
def _split_generators(self, dl_manager: ds.DownloadManager):
|
534 |
if self.config.name == "MARC-ja":
|
535 |
+
return self.__split_generators_marc_ja(dl_manager)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
536 |
else:
|
537 |
+
return self.__split_generators(dl_manager)
|
538 |
+
|
539 |
+
def __generate_examples_marc_ja(self, split_df: Optional[pd.DataFrame] = None):
|
540 |
+
if split_df is None:
|
541 |
+
raise ValueError(f"Invalid preprocessing for {self.config.name}")
|
542 |
+
|
543 |
+
instances = split_df.to_dict(orient="records")
|
544 |
+
for i, data_dict in enumerate(instances):
|
545 |
+
yield i, data_dict
|
546 |
+
|
547 |
+
def __generate_examples_jsquad(self, file_path: Optional[str] = None):
|
548 |
+
if file_path is None:
|
549 |
+
raise ValueError(f"Invalid argument for {self.config.name}")
|
550 |
+
|
551 |
+
with open(file_path, "r") as rf:
|
552 |
+
json_data = json.load(rf)
|
553 |
+
|
554 |
+
for json_dict in json_data["data"]:
|
555 |
+
title = json_dict["title"]
|
556 |
+
paragraphs = json_dict["paragraphs"]
|
557 |
+
|
558 |
+
for paragraph in paragraphs:
|
559 |
+
context = paragraph["context"]
|
560 |
+
questions = paragraph["qas"]
|
561 |
+
|
562 |
+
for question_dict in questions:
|
563 |
+
q_id = question_dict["id"]
|
564 |
+
question = question_dict["question"]
|
565 |
+
answers = question_dict["answers"]
|
566 |
+
is_impossible = question_dict["is_impossible"]
|
567 |
+
|
568 |
+
example_dict = {
|
569 |
+
"id": q_id,
|
570 |
+
"title": title,
|
571 |
+
"context": context,
|
572 |
+
"question": question,
|
573 |
+
"answers": answers,
|
574 |
+
"is_impossible": is_impossible,
|
575 |
+
}
|
576 |
+
|
577 |
+
yield q_id, example_dict
|
578 |
+
|
579 |
+
def __generate_examples_jcommonsenseqa(self, file_path: Optional[str] = None):
|
580 |
+
if file_path is None:
|
581 |
+
raise ValueError(f"Invalid argument for {self.config.name}")
|
582 |
+
|
583 |
+
with open(file_path, "r") as rf:
|
584 |
+
for i, line in enumerate(rf):
|
585 |
+
json_dict = json.loads(line)
|
586 |
+
json_dict["label"] = f"choice{json_dict['label']}"
|
587 |
+
yield i, json_dict
|
588 |
+
|
589 |
+
def __generate_examples(self, file_path: Optional[str] = None):
|
590 |
+
if file_path is None:
|
591 |
+
raise ValueError(f"Invalid argument for {self.config.name}")
|
592 |
+
|
593 |
+
with open(file_path, "r") as rf:
|
594 |
+
for i, line in enumerate(rf):
|
595 |
+
json_dict = json.loads(line)
|
596 |
+
yield i, json_dict
|
597 |
|
598 |
def _generate_examples(
|
599 |
self,
|
|
|
601 |
split_df: Optional[pd.DataFrame] = None,
|
602 |
):
|
603 |
if self.config.name == "MARC-ja":
|
604 |
+
yield from self.__generate_examples_marc_ja(split_df)
|
|
|
605 |
|
606 |
+
elif self.config.name == "JSQuAD":
|
607 |
+
yield from self.__generate_examples_jsquad(file_path)
|
608 |
+
|
609 |
+
elif self.config.name == "JCommonsenseQA":
|
610 |
+
yield from self.__generate_examples_jcommonsenseqa(file_path)
|
611 |
|
612 |
else:
|
613 |
+
yield from self.__generate_examples(file_path)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|