wwu-mmll / confound_corrected_cpm / 28544354110

85%

main: 85%

push

github

Fix edge selection running on CPU when GPU requested; drop dead correlation helpers

_select_edges called fit_transform without passing device, so the
default (CPU) was always used even when device='gpu'/'cuda'. Edge
selection — often the heaviest part under permutations — therefore never
ran on the GPU. Allocate the edges accumulator on self.device and thread
device=self.device into fit_transform.

Also remove the misleading module-level `device` global in
edge_selection.py (never read; contradicts the explicit device-passing
design) and the unused numpy correlation helpers
(compute_correlation_and_pvalues and friends, one_sample_t_test) plus
their imports and the two tests that only exercised them. The production
path (correlations_and_pvalues) stays covered.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

1 of 1 new or added line in 1 file covered. (100.0%)

2110 of 2913 relevant lines covered (72.43%)

0.72 hits per line