import csv import datasets _TRAIN_DOWNLOAD_URL = "https://drive.google.com/u/0/uc?id=1RWSlXKGqjx9KshiOnm-HxQxJVVlk3TmV&export=download" _DEV_DOWNLOAD_URL = "https://drive.google.com/u/0/uc?id=13ZGbKuJFXWjHkkPSkKgJoLuklZpuPipE&export=download" _TEST_DOWNLOAD_URL = "https://drive.google.com/u/0/uc?id=1YrnAmkHC9Mgnn99azSpmAKRwMxTViIKC&export=download" class Hotel(datasets.GeneratorBasedBuilder): def _info(self): return datasets.DatasetInfo( description=None, features=datasets.Features( { "text": datasets.Value("string"), "label": datasets.features.ClassLabel(names=["negative", "positive"]), } ), homepage=None, citation=None, ) def _split_generators(self, dl_manager): train_path = dl_manager.download_and_extract(_TRAIN_DOWNLOAD_URL) dev_path = dl_manager.download_and_extract(_DEV_DOWNLOAD_URL) test_path = dl_manager.download_and_extract(_TEST_DOWNLOAD_URL) return [ datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": train_path}), datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": dev_path}), datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": test_path}), ] def _generate_examples(self, filepath): with open(filepath, encoding="utf-8") as csv_file: csv_reader = csv.reader( csv_file, quotechar='"', delimiter=",", quoting=csv.QUOTE_ALL, skipinitialspace=True ) for id_, row in enumerate(csv_reader): if id_ == 0: continue label, text = row label = int(label) yield id_, {"text": text, "label": label}