newsph_nli

参照:

次のコマンドを使用して、このデータセットをTFDSにロードします。

ds = tfds.load('huggingface:newsph_nli')
  • 説明
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.
  • ライセンス:フィリピン語-テキスト-ベンチマークはGNU General PublicLicensev3.0の下でライセンスされています
  • バージョン:1.0.0
  • 分割
スプリット
'test' 9000
'train' 420000
'validation' 90000
  • 特徴
{
    "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"
    }
}