newsph_nli

참조:

다음 명령을 사용하여 TFDS에서 이 데이터세트를 로드합니다.

ds = tfds.load('huggingface:newsph_nli')
  • 설명 :
First benchmark dataset for sentence entailment in the low-resource Filipino language. Constructed through exploting the structure of news articles. Contains 600,000 premise-hypothesis pairs, in 70-15-15 split for training, validation, and testing.
  • 라이선스 : Filipino-Text-Benchmarks는 GNU General Public License v3.0에 따라 라이선스가 부여되었습니다.
  • 버전 : 1.0.0
  • 분할 :
나뉘다
'test' 9000
'train' 420000
'validation' 90000
  • 특징 :
{
    "premise": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "hypothesis": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "label": {
        "num_classes": 2,
        "names": [
            "0",
            "1"
        ],
        "names_file": null,
        "id": null,
        "_type": "ClassLabel"
    }
}