Metric Function#
braintools.metric collects losses and metrics for training models. The losses are pure JAX functions that match their optax counterparts.
The measures for analysing neural activity (firing rates, spike-train synchrony and distances, LFP analysis, functional connectivity) moved to braintools.analysis in 0.4; their tutorials are now under Neural Data Analysis. The old braintools.metric names still work but warn when called and will be removed in 1.0.
What you’ll find here#
Classification losses: sigmoid and softmax cross-entropy, focal loss, label smoothing, and a gradient-descent example.
Regression losses: L1, L2, Huber, and log-cosh; the differing default reductions; and how loss shape controls robustness to outliers.
Ranking: the listwise softmax loss with masking, weights, and reductions, and a small learning-to-rank example.
Cosine similarity: aligned pairs versus pairwise matrices, retrieval, and comparing population activity patterns.
For the full list of functions and parameters, see the API reference.