htfa.stats

Reconstruction scores and nominal conditional Gaussian information criteria.

These are descriptive plug-in criteria, not marginal evidence for HTFA. Spatial and temporal residual dependence, regularization and mixture singularities break the usual independent-observation / regular maximum-likelihood justification.

Functions

gaussian_fit_statistics(observed, predicted, ...)	Score equally shaped (voxels, times) arrays or lists of subject arrays.
parameter_count(results)	Nominal conditional reconstruction parameter budget for fitted TFA/HTFA.
reconstruction_statistics(results)	Score full fitted subjects in the model's standardized fitting space.

Classes

FitStatistics(n_observations, n_parameters, ...)

Pooled scores; smaller information criteria and larger R² are better.