"""Korean Balanced Evaluation of Significant Tasks""" import csv import os import pandas as pd import datasets _CITATAION = """\ @misc{https://doi.org/10.48550/arxiv.2204.04541, doi = {10.48550/ARXIV.2204.04541}, url = {https://arxiv.org/abs/2204.04541}, author = {Kim, Dohyeong and Jang, Myeongjun and Kwon, Deuk Sin and Davis, Eric}, title = {KOBEST: Korean Balanced Evaluation of Significant Tasks}, publisher = {arXiv}, year = {2022}, } """ _DESCRIPTION = """\ The dataset contains data for KoBEST dataset """ _URL = "https://github.com/SKT-LSL/KoBEST_datarepo/raw/main" _DATA_URLS = { "boolq": { "train": _URL + "/v1.0/BoolQ/train.tsv", "dev": _URL + "/v1.0/BoolQ/dev.tsv", "test": _URL + "/v1.0/BoolQ/test.tsv", }, "copa": { "train": _URL + "/v1.0/COPA/train.tsv", "dev": _URL + "/v1.0/COPA/dev.tsv", "test": _URL + "/v1.0/COPA/test.tsv", }, "sentineg": { "train": _URL + "/v1.0/SentiNeg/train.tsv", "dev": _URL + "/v1.0/SentiNeg/dev.tsv", "test": _URL + "/v1.0/SentiNeg/test.tsv", "test_originated": _URL + "/v1.0/SentiNeg/test.tsv", }, "hellaswag": { "train": _URL + "/v1.0/HellaSwag/train.tsv", "dev": _URL + "/v1.0/HellaSwag/dev.tsv", "test": _URL + "/v1.0/HellaSwag/test.tsv", }, "wic": { "train": _URL + "/v1.0/WiC/train.tsv", "dev": _URL + "/v1.0/WiC/dev.tsv", "test": _URL + "/v1.0/WiC/test.tsv", }, } _LICENSE = "CC-BY-SA-4.0" class KoBESTConfig(datasets.BuilderConfig): """Config for building KoBEST""" def __init__(self, description, data_url, citation, url, **kwargs): """ Args: description: `string`, brief description of the dataset data_url: `dictionary`, dict with url for each split of data. citation: `string`, citation for the dataset. url: `string`, url for information about the dataset. **kwrags: keyword arguments frowarded to super """ super(KoBESTConfig, self).__init__(version=datasets.Version("1.0.0", ""), **kwargs) self.description = description self.data_url = data_url self.citation = citation self.url = url class KoBEST(datasets.GeneratorBasedBuilder): BUILDER_CONFIGS = [ KoBESTConfig(name=name, description=_DESCRIPTION, data_url=_DATA_URLS[name], citation=_CITATAION, url=_URL) for name in ["boolq", "copa", 'sentineg', 'hellaswag', 'wic'] ] BUILDER_CONFIG_CLASS = KoBESTConfig def _info(self): features = {} if self.config.name == "boolq": labels = ["False", "True"] features["paragraph"] = datasets.Value("string") features["question"] = datasets.Value("string") features["label"] = datasets.features.ClassLabel(names=labels) if self.config.name == "copa": labels = ["alternative_1", "alternative_2"] features["premise"] = datasets.Value("string") features["question"] = datasets.Value("string") features["alternative_1"] = datasets.Value("string") features["alternative_2"] = datasets.Value("string") features["label"] = datasets.features.ClassLabel(names=labels) if self.config.name == "wic": labels = ["False", "True"] features["word"] = datasets.Value("string") features["context_1"] = datasets.Value("string") features["context_2"] = datasets.Value("string") features["label"] = datasets.features.ClassLabel(names=labels) if self.config.name == "hellaswag": labels = ["ending_1", "ending_2", "ending_3", "ending_4"] features["context"] = datasets.Value("string") features["ending_1"] = datasets.Value("string") features["ending_2"] = datasets.Value("string") features["ending_3"] = datasets.Value("string") features["ending_4"] = datasets.Value("string") features["label"] = datasets.features.ClassLabel(names=labels) if self.config.name == "sentineg": labels = ["negative", "positive"] features["sentence"] = datasets.Value("string") features["label"] = datasets.features.ClassLabel(names=labels) return datasets.DatasetInfo( description=_DESCRIPTION, features=datasets.Features(features), homepage=_URL, citation=_CITATAION ) def _split_generators(self, dl_manager): train = dl_manager.download_and_extract(self.config.data_url["train"]) dev = dl_manager.download_and_extract(self.config.data_url["dev"]) test = dl_manager.download_and_extract(self.config.data_url["test"]) if self.config.data_url.get("test_originated"): test_originated = dl_manager.download_and_extract(self.config.data_url["test_originated"]) return [ datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": train, "split": "train"}), datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": dev, "split": "dev"}), datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": test, "split": "test"}), datasets.SplitGenerator(name="test_originated", gen_kwargs={"filepath": test_originated, "split": "test_originated"}), ] return [ datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": train, "split": "train"}), datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": dev, "split": "dev"}), datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": test, "split": "test"}), ] def _generate_examples(self, filepath, split): if self.config.name == "boolq": df = pd.read_csv(filepath, sep="\t") df = df.dropna() df = df[['Text', 'Question', 'Answer']] df = df.rename(columns={ 'Text': 'paragraph', 'Question': 'question', 'Answer': 'label', }) df['label'] = [0 if str(s) == 'False' else 1 for s in df['label'].tolist()] elif self.config.name == "copa": df = pd.read_csv(filepath, sep="\t") df = df.dropna() df = df[['sentence', 'question', '1', '2', 'Answer']] df = df.rename(columns={ 'sentence': 'premise', 'question': 'question', '1': 'alternative_1', '2': 'alternative_2', 'Answer': 'label', }) df['label'] = [i-1 for i in df['label'].tolist()] elif self.config.name == "wic": df = pd.read_csv(filepath, sep="\t") df = df.dropna() df = df[['Target', 'SENTENCE1', 'SENTENCE2', 'ANSWER']] df = df.rename(columns={ 'Target': 'word', 'SENTENCE1': 'context_1', 'SENTENCE2': 'context_2', 'ANSWER': 'label', }) df['label'] = [0 if str(s) == 'False' else 1 for s in df['label'].tolist()] elif self.config.name == "hellaswag": df = pd.read_csv(filepath, sep="\t") df = df.dropna() df = df[['context', 'choice1', 'choice2', 'choice3', 'choice4', 'label']] df = df.rename(columns={ 'context': 'context', 'choice1': 'ending_1', 'choice2': 'ending_2', 'choice3': 'ending_3', 'choice4': 'ending_4', 'label': 'label', }) elif self.config.name == "sentineg": df = pd.read_csv(filepath, sep="\t") df = df.dropna() if split == "test_originated": df = df[['Text_origin', 'Label_origin']] df = df.rename(columns={ 'Text_origin': 'sentence', 'Label_origin': 'label', }) else: df = df[['Text', 'Label']] df = df.rename(columns={ 'Text': 'sentence', 'Label': 'label', }) else: raise NotImplementedError for id_, row in df.iterrows(): features = {key: row[key] for key in row.keys()} yield id_, features if __name__ == "__main__": for config_name in ["boolq", "copa", 'sentineg', 'hellaswag', 'wic']: dataset = datasets.load_dataset("kobest_v1.py", config_name, ignore_verifications=True) os.makedirs(config_name, exist_ok=True) for split, split_dataset in dataset.items(): split_dataset.to_json(f"{config_name}/{split}.jsonl") # for task in ['boolq', 'copa', 'wic', 'hellaswag', 'sentineg']: # dataset = datasets.load_dataset("kobest_v1.py", task, ignore_verifications=True) # print(dataset) # print(dataset['train']['label'])