wwu-mmll / confound_corrected_cpm / 35611611867

88%

main: 88%

Build Type

push

github

Committed by NilsWinter

Commit Message

Phase 7: cut 29 redundant tests, add two that check something new

The principle, agreed up front: keep every test that checks against
sklearn/scipy/numpy or an analytically known ground truth; delete tests that
only restate an implementation detail or duplicate a reference check with
hand-computed numbers.

Cuts (29 collected items):

  test_scoring          -10  4x perfect_predictions_* and 2x *_calculation are
                             hand-computed restatements of the vs-sklearn
                             checks; negative-correlation would fail
                             test_pearson_vs_scipy first; plus an initializer
                             assertion and a "returns a tensor" assertion
  test_connectome        -4  three tests round-tripped the test-local NumPy
                             reference against itself, not production code;
                             one checked only a shape the next test checks
                             elementwise
  test_results_manager   -4  constructor-stored-its-arguments, a store_edges
                             test subsumed by the stability test that stores
                             edges, "agg_results is a DataFrame", and one of
                             two edge-tensor shape tests (merged, keeping the
                             dtype assertion)
  test_simulate_sem      -5  isinstance/shape smoke and two dict-key
                             restatements; the statistical-property tests stay
  test_simulate_simple   -4  shape smoke and an isinstance check over three
                             link types
  test_inference         -2  mixed_metrics is the union of the two
                             higher/lower-is-better tests; nbs p-value bounds
                             merged into the NBS contract test

Evidence the cuts were redundant: line coverage is **unchanged** on every
affected module -- connectome 100%, inference 100%, scoring 99%, results_manager
89%, simulate_sem 87%,... (continued)

Coverage Stats

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

94 existing lines in 4 files now uncovered.

2415 of 2774 relevant lines covered (87.06%)

0.87 hits per line