quaterion.distances.dot_product module

class DotProduct

Bases: BaseDistance

Compute dot product similarities (and its interpretation as distances).

Warning

Interpretation of dot product as distances may have unexpected effects. Make sure that you entirely understand how it exactly works, and when combined and with the chosen loss function in particular, because those values are negative.

static distance(x: Tensor, y: Tensor) → Tensor

Calculate distances, i.e., the lower the value, the more similar the samples.

Parameters:

• x – shape: (batch_size, embedding_dim)

• y – shape: (batch_size, embedding_dim)

Returns:

Distances - shape – (batch_size,)

static distance_matrix(x: Tensor, y: Tensor | None = None) → Tensor

Calculate a distance matrix, i.e., distances between all possible pairs in x and y.

Parameters:

• x – shape: (batch_size, embedding_dim)

• y – shape: (batch_size, embedding_dim). If y is None, it assigns x to y.

Returns:

Distance matrix - shape – (batch_size, batch_size)

static similarity(x: Tensor, y: Tensor) → Tensor

Calculate similarities, i.e., the higher the value, the more similar the samples.

Parameters:

• x – shape: (batch_size, embedding_dim)

• y – shape: (batch_size, embedding_dim)

Returns:

Similarities - shape – (batch_size,)

static similarity_matrix(x: Tensor, y: Tensor | None = None) → Tensor

Calculate a similarity matrix, i.e., similarities between all possible pairs in x and y.

Parameters:

• x – shape: (batch_size, embedding_dim)

• y – shape: (batch_size, embedding_dim). If y is None, it assigns x to y.

Returns:

Similarity matrix - shape – (batch_size, batch_size)