skada.deep.losses.nap_loss

skada.deep.losses.nap_loss(features_t, y_pred_t, memory_features, memory_outputs, sample_idx_t, K=5)

Compute the NAP loss.

Inspired by https://github.com/CrownX/SPA

Parameters:

features_ttorch.Tensor

Target features.

y_pred_ttorch.Tensor

Target predictions.

memory_featurestorch.Tensor

Memory features.

memory_outputstorch.Tensor

Memory outputs.

sample_idx_ttorch.Tensor

The sample indices in the batch features_t

Kint, default=5

The number of nearest neighbors.