ListMLELoss

[source]

ListMLELoss class

keras_rs.losses.ListMLELoss(temperature: float = 1.0, **kwargs: Any)

Implements ListMLE (Maximum Likelihood Estimation) loss for ranking.

ListMLE loss is a listwise ranking loss that maximizes the likelihood of the ground truth ranking. It works by: 1. Sorting items by their relevance scores (labels) 2. Computing the probability of observing this ranking given the predicted scores 3. Maximizing this likelihood (minimizing negative log-likelihood)

The loss is computed as the negative log-likelihood of the ground truth ranking given the predicted scores:

loss = -sum(log(exp(s_i) / sum(exp(s_j) for j >= i)))

where s_i is the predicted score for item i in the sorted order.

Arguments

• temperature: Temperature parameter for scaling logits. Higher values make the probability distribution more uniform. Defaults to 1.0.
• reduction: Type of reduction to apply to the loss. In almost all cases this should be "sum_over_batch_size". Supported options are "sum", "sum_over_batch_size", "mean", "mean_with_sample_weight" or None. Defaults to "sum_over_batch_size".
• name: Optional name for the loss instance.
• dtype: The dtype of the loss's computations. Defaults to None.

Examples

# Basic usage
loss_fn = ListMLELoss()

# With temperature scaling
loss_fn = ListMLELoss(temperature=0.5)

# Example with synthetic data
y_true = [[3, 2, 1, 0]]  # Relevance scores
y_pred = [[0.8, 0.6, 0.4, 0.2]]  # Predicted scores
loss = loss_fn(y_true, y_pred)

[source]

call method

ListMLELoss.call(y_true: Any, y_pred: Any)

Compute the ListMLE loss.

Arguments

• y_true: tensor or dict. Ground truth values. If tensor, of shape (list_size) for unbatched inputs or (batch_size, list_size) for batched inputs. If an item has a label of -1, it is ignored in loss computation. If it is a dictionary, it should have two keys: "labels" and "mask". "mask" can be used to ignore elements in loss computation.
• y_pred: tensor. The predicted values, of shape (list_size) for unbatched inputs or (batch_size, list_size) for batched inputs. Should be of the same shape as y_true.

Returns

The loss tensor of shape [batch_size].