quaterion.loss.group_loss module

class GroupLoss(distance_metric_name: Distance = Distance.COSINE)[source]

Bases: SimilarityLoss

Base class for group losses.

Parameters:

distance_metric_name – Name of the distance function, e.g., Distance.

forward(embeddings: Tensor, groups: LongTensor) → Tensor[source]

Parameters:

• embeddings – shape: (batch_size, vector_length)

• groups – shape: (batch_size,) - Groups, associated with embeddings

Returns:

Tensor – zero-size tensor, loss value

training: bool