loss

fairdiverse.search.utils.loss.list_pairwise_loss(score_1, score_2, delta)[source]

Computes the accuracy and list-pairwise loss based on the difference between two scores.

Parameters:

  • score_1 – A tensor containing the scores for the first set of items.

  • score_2 – A tensor containing the scores for the second set of items.

  • delta – A tensor representing the weight for each pairwise comparison.

Returns:

The accuracy and the computed list-pairwise loss.

fairdiverse.search.utils.loss.ndcg_loss(score, div)[source]

Computes the NDCG (Normalized Discounted Cumulative Gain) loss for ranking tasks.

Parameters:

  • score – A tensor containing the predicted scores for each document.

  • div – A tensor containing the relevance or diversity values for each document.

Returns:

The computed NDCG loss.