Skip to contents

Common per-batch loss measures for TaskLCE / LearnerLCE evaluations. Each measure aggregates predictions and truth to one value per batch (via lce_per_batch) and then computes a standard regression loss on those per-batch pairs.

  • lce.mse: mean squared error.

  • lce.rmse: root mean squared error (the square root of the per-batch MSE).

  • lce.mae: mean absolute error.

All measures support observation weights from a weights_measure task column; the weight applied to a batch is the sum of weights of its archive rows.