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.