sift1m

  • 説明:

ユークリッド距離を使用した近似最近傍検索用の事前トレーニング済み埋め込み。このデータセットは、次の 2 つの分割で構成されています。

  1. 'database': 1,000,000 個のデータ ポイントで構成され、それぞれに次の機能があります: 'embedding' (128 float)、'index' (int64)、'neighbors' (空のリスト)。
  2. 'test': 10,000 個のデータ ポイントで構成され、それぞれに次の機能があります: 'embedding' (128 float)、'index' (int64)、'neighbors' (データベース内の最近傍の 'index' と 'distance' のリスト)。 )
スプリット
'database' 1,000,000
'test' 10,000
  • 機能構造:
FeaturesDict({
    'embedding': Tensor(shape=(128,), dtype=float32),
    'index': Scalar(shape=(), dtype=int64),
    'neighbors': Sequence({
        'distance': Scalar(shape=(), dtype=float32),
        'index': Scalar(shape=(), dtype=int64),
    }),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
埋め込みテンソル(128) float32
索引スカラーint64分割内のインデックス。
隣人順序テスト分割でのみ使用できる、計算された隣接ノード。
隣人/距離スカラーfloat32近隣距離。
ネイバー/インデックススカラーint64ネイバー インデックス。
  • 引用
@article{jegou2010product,
  title={Product quantization for nearest neighbor search},
  author={Jegou, Herve and Douze, Matthijs and Schmid, Cordelia},
  journal={IEEE transactions on pattern analysis and machine intelligence},
  volume={33},
  number={1},
  pages={117--128},
  year={2010},
  publisher={IEEE}
}