[ChrisW09]

Fixes #508

Problem

NegativeLogLikelihood.__call__ turned y_true into a tensor unchanged and called distribution.compute_loss(y_pred, y_true). For the (N, 1) targets that the training loop passes (and that evaluate() forwards unchanged when y_true is a column vector), log_prob broadcast the batch of N distributions against the column vector into an (N, N) grid, and the metric returned the mean of that grid. A val_metrics={"nll": NegativeLogLikelihood(family)} therefore logged a different value from val_loss for the same family. LogScore wraps the same metric and had the same problem.

Fix

NegativeLogLikelihood squeezes a trailing singleton dimension of y_true ((N, 1) → (N,)) before computing the loss. Multi-output targets (e.g. Dirichlet components) and 1-D targets are unchanged.

Tests

tests/metrics/test_metric_contracts.py::TestLikelihoodMetricsAcceptColumnTargets:

• test_column_and_flat_targets_give_the_family_loss[NegativeLogLikelihood|LogScore]: flat and column targets both give the family's compute_loss (negated for LogScore).
• test_logged_validation_nll_matches_validation_loss: in an MLPLSS fit, the logged val_nll equals val_loss.

All three fail on main and pass with this change.

Docs

No change needed. The metric was already documented as the family NLL.

Checks

• just test: 3239 passed, 46 skipped, 28 xfailed; coverage 91.38%
• just docs: build succeeded (warnings treated as errors)
• just check: all hooks passed (ruff, prettier, pyright)