wwu-mmll / confound_corrected_cpm / 35608509104

88%

main: 88%

Build Type

push

github

Committed by NilsWinter

Commit Message

Split PermutationManager into inference.py and the edge statistics into statistics.py

Two files each held two unrelated things.

results_manager.py (737 lines) mixed an accumulator with ~360 lines of
permutation statistics. PermutationManager moves to inference.py -- NBS, TFCE
and the permutation p-values, the most statistically delicate code in the
package, now findable under a name that says what it is. ResultsManager keeps
nothing more exotic than means and argmax. 737 -> 381 + 388.

edge_selection.py (595 lines) mixed the maths with the policy built on it.
torch_rankdata, get_residuals, torch_bonferroni and correlations_and_pvalues
move to statistics.py; the selectors, the presence and connected-component
filters, and the UnivariateEdgeSelection facade stay. 595 -> 385 + 234.

Tests follow the code: tests/test_inference.py and tests/test_statistics.py,
carved out of test_results_manager.py and test_edge_selection.py. The same
tests, none dropped, plus two dead _uneven_splits helpers removed.

One bug found and fixed in the process: inference.py calls
ResultsManager.load_cv_results, which was a module global in the old file and
became an unimported name in the new one. The whole unit suite passed --
only test_integration.py caught it, via the example scripts. So
calculate_permutation_results, the top-level entry point of this module, now
has a direct test that builds a minimal results directory on disk and runs both
methods in ~1s. Verified that it fails with the import removed.

Suite 317 -> 326.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Coverage Stats

220 of 222 new or added lines in 5 files covered. (99.1%)

32 existing lines in 2 files now uncovered.

2418 of 2778 relevant lines covered (87.04%)

0.87 hits per line