fusionlab.nn.losses.prediction_based_loss

fusionlab.nn.losses.prediction_based_loss(quantiles=None, anomaly_loss_weight=0.1)

Create a combined prediction + anomaly loss function for prediction-based strategy.

Parameters:

• quantiles (list of float, optional) – Quantiles for quantile loss calculation. If None, uses MSE.

• anomaly_loss_weight (float, default 0.1) – Weight for anomaly loss component.

Returns:

A loss function: loss_fn(y_true, y_pred)

Return type:

callable

Notes

• Handles both quantile and MSE-based prediction losses

• Anomaly loss is computed as mean tf_squared prediction errors

• Compatible with Keras serialization/deserialization