quaterion.dataset.similarity_samples 模块

class SimilarityGroupSample(obj: Any, group: int)

基类: object

表示相似对象组，组内的所有对象应相互匹配。

示例

Faces dataset.
All pictures of a single person should have single unique group
id. In this case NN will learn to match all pictures within
the group closer to each-other, but pictures from different
groups - further.

             file_name  group_id
0      elon_musk_1.jpg       555
1      elon_musk_2.jpg       555
2      elon_musk_3.jpg       555
3  leonard_nimoy_1.jpg       209
4  leonard_nimoy_2.jpg       209

group: int

obj: Any

class SimilarityPairSample(obj_a: Any, obj_b: Any, score: float = 1.0, subgroup: int = 0)

基类: object

表示一对对象、它们的相似性以及与其他对的关系。

示例

data = [
    # First query group (subgroup)
    SimilarityPairSample(
        obj_a="cheesecake",
        obj_b="muffins",
        score=0.9,
        subgroup=10
    ),
    SimilarityPairSample(
        obj_a="cheesecake",
        obj_b="macaroons",
        score=0.8,
        subgroup=10
    ),
    SimilarityPairSample(
        obj_a="cheesecake",
        obj_b="candies",
        score=0.7,
        subgroup=10
    ),

    # Second query group (subgroup)
    SimilarityPairSample(
        obj_a="lemon",
        obj_b="lime",
        score=0.9,
        subgroup=11
    ),
    SimilarityPairSample(
        obj_a="lemon",
        obj_b="orange",
        score=0.7,
        subgroup=11
    ),
]

obj_a: Any

obj_b: Any

score: float = 1.0

对象 A 和 B 之间的相似性得分。假定 score = 1.0 表示对象相似，score = 0.0 表示对象完全不同。

subgroup: int =0

将此组之外的所有示例视为负样本。默认情况下，所有样本都属于组 0 - 因此其他样本不能用作负示例。