nli_zh / nli_zh.py
shibing624's picture
Update nli_zh.py
2a8a4cb
raw
history blame
4.92 kB
# -*- coding: utf-8 -*-
"""
@author:XuMing(xuming624@qq.com)
@description:
"""
"""Natural Language Inference (NLI) Chinese Corpus.(nli_zh)"""
import os
import datasets
_DESCRIPTION = """纯文本数据,格式:(sentence1, sentence2, label)。常见中文语义匹配数据集,包含ATEC、BQ、LCQMC、PAWSX、STS-B共5个任务。"""
ATEC_HOME = "https://github.com/IceFlameWorm/NLP_Datasets/tree/master/ATEC"
BQ_HOME = "http://icrc.hitsz.edu.cn/info/1037/1162.htm"
LCQMC_HOME = "http://icrc.hitsz.edu.cn/Article/show/171.html"
PAWSX_HOME = "https://arxiv.org/abs/1908.11828"
STSB_HOME = "https://github.com/pluto-junzeng/CNSD"
_CITATION = "https://github.com/shibing624/text2vec"
_DATA_URL = "https://github.com/shibing624/text2vec/releases/download/1.1.2/senteval_cn.zip"
class NliZhConfig(datasets.BuilderConfig):
"""BuilderConfig for NLI_zh"""
def __init__(self, features, data_url, citation, url, label_classes=(0, 1), **kwargs):
"""BuilderConfig for NLI_zh
Args:
features: `list[string]`, list of the features that will appear in the
feature dict. Should not include "label".
data_url: `string`, url to download the zip file from.
citation: `string`, citation for the data set.
url: `string`, url for information about the data set.
label_classes: `list[int]`, sim is 1, else 0.
**kwargs: keyword arguments forwarded to super.
"""
super().__init__(version=datasets.Version("1.0.0"), **kwargs)
self.features = features
self.label_classes = label_classes
self.data_url = data_url
self.citation = citation
self.url = url
class NliZh(datasets.GeneratorBasedBuilder):
"""The Natural Language Inference Chinese(NLI_zh) Corpus."""
BUILDER_CONFIGS = [
NliZhConfig(
name="ATEC",
description=_DESCRIPTION,
features=["sentence1", "sentence1"],
data_url=_DATA_URL,
citation=_CITATION,
url=ATEC_HOME,
),
NliZhConfig(
name="BQ",
description=_DESCRIPTION,
features=["sentence1", "sentence1"],
data_url=_DATA_URL,
citation=_CITATION,
url=BQ_HOME,
),
NliZhConfig(
name="LCQMC",
description=_DESCRIPTION,
features=["sentence1", "sentence1"],
data_url=_DATA_URL,
citation=_CITATION,
url=LCQMC_HOME,
),
NliZhConfig(
name="PAWSX",
description=_DESCRIPTION,
features=["sentence1", "sentence1"],
data_url=_DATA_URL,
citation=_CITATION,
url=PAWSX_HOME,
),
NliZhConfig(
name="STS-B",
description=_DESCRIPTION,
features=["sentence1", "sentence1"],
data_url=_DATA_URL,
citation=_CITATION,
url=STSB_HOME,
),
]
def _info(self):
return datasets.DatasetInfo(
description=self.config.description,
features=datasets.Features(
{
"sentence1": datasets.Value("string"),
"sentence2": datasets.Value("string"),
"label": datasets.Value("int32"),
# "idx": datasets.Value("int32"),
}
),
homepage=self.config.url,
citation=self.config.citation,
)
def _split_generators(self, dl_manager):
dl_dir = dl_manager.download_and_extract(self.config.data_url) or ""
dl_dir = os.path.join(dl_dir, f"senteval_cn/{self.config.name}")
return [
datasets.SplitGenerator(
name=datasets.Split.TRAIN,
gen_kwargs={
"filepath": os.path.join(dl_dir, f"{self.config.name}.train.data"),
},
),
datasets.SplitGenerator(
name=datasets.Split.VALIDATION,
gen_kwargs={
"filepath": os.path.join(dl_dir, f"{self.config.name}.valid.data"),
},
),
datasets.SplitGenerator(
name=datasets.Split.TEST,
gen_kwargs={
"filepath": os.path.join(dl_dir, f"{self.config.name}.test.data"),
},
),
]
def _generate_examples(self, filepath):
"""This function returns the examples in the raw (text) form."""
with open(filepath, 'r', encoding="utf-8") as f:
for idx, row in enumerate(f):
# print(row)
terms = row.split('\t')
yield idx, {
"sentence1": terms[0],
"sentence2": terms[1],
"label": int(terms[2]),
}