similarity

fgvc.special.similarity.batch_cosine_similarity(a: Tensor | ndarray, b: Tensor | ndarray, eps: float = 1e-08) → Tensor | ndarray

Compute batch-wise cosine similarity between a vector and a vector or matrix.

Parameters:

• a – A batch of vectors represented as numpy array or pytorch tensor.

• b – A batch of vectors or matrices represented as numpy array or pytorch tensor.

• eps – Epsilon small constant to prevent division by 0.

Returns:

Cosine similarity vector or matrix.

Return type:

cos_sim

fgvc.special.similarity.cosine_similarity(a: Tensor | ndarray, b: Tensor | ndarray, eps: float = 1e-08) → Tensor | ndarray | float

Compute cosine similarity between a vector and a vector or matrix.

Parameters:

• a – A vector represented as numpy array or pytorch tensor.

• b – A vector or a matrix represented as numpy array or pytorch tensor.

• eps – Epsilon small constant to prevent division by 0.

Returns:

Cosine similarity number or vector.

Return type:

cos_sim

fgvc.special.similarity.pairwise_cosine_similarity(mat: Tensor | ndarray, eps: float = 1e-08, progress: bool = False) → Tensor | ndarray

Compute pair-wise cosine similarity between all elements in the input matrix.

Parameters:

• mat – A matrix represented as numpy array or pytorch tensor.

• eps – Epsilon small constant to prevent division by 0.

• progress – If true use tqdm progress bar.

Returns:

A similarity matrix.

Return type:

sim_mat