SFI.inference.sparse.metrics module

SFI.inference.sparse.metrics — Benchmark helpers

Standalone functions for comparing inferred supports / coefficients against ground truth. Useful for benchmarking and papers but not required for normal inference.

SFI.inference.sparse.metrics.overlap_metrics(true_support, pred_support)[source]

Compare predicted support to the ground truth.

Parameters:

  • true_support (list[int]) – Indices of the true and predicted active basis functions.

  • pred_support (list[int]) – Indices of the true and predicted active basis functions.

Returns:

Keys: TP, FP, FN, prec, rec, exact.

Return type:

dict

SFI.inference.sparse.metrics.predictive_nmse(Phi_test, true_support, true_coeffs, inferred_support, inferred_coeffs)[source]

Normalised mean-squared error on a held-out design matrix.

Parameters:

  • Phi_test ((n_test, p) Array) – Design matrix evaluated on test data.

  • true_support (list[int]) – Ground-truth active indices.

  • true_coeffs (array-like) – Ground-truth coefficient vector (length len(true_support)).

  • inferred_support (list[int]) – Inferred active indices.

  • inferred_coeffs (array-like) – Inferred coefficient vector (length len(inferred_support)).

Returns:

\(\|\hat y - y\|^2 / \|y\|^2\).

Return type:

float