daniel1noble / BACE / 25334172491
91%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 15
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 76.022% (+4.6%) from 71.444%
25334172491

push

github

daniel1noble 
Add per-type imputation accuracy tests + pipeline numerical assertions

Adds two more test files driving toward the ambitious 1000-test /
90%-coverage target. Local: 570 PASS / 0 FAIL.

tests/testthat/test-imputation_accuracy.R (6 tests, all skip_on_cran):
End-to-end accuracy checks across every BACE response-type path.
Each simulates ground truth, masks 30%, runs bace_imp, and asserts
NUMERICAL output properties (not just shape):

  continuous : Pearson r(imp, truth) > 0.30; bace_rmse < 1.05 *
               column-mean baseline (5% slack absorbs small-N noise).
  count      : imputed values non-negative integers; finite.
  binary     : BACE accuracy >= modal-class baseline accuracy.
  ordinal    : mean abs level distance < 2.0 (better than uniform).
  categorical: BACE accuracy > random + within 20pp of modal-class
               baseline (catches catastrophic regressions).
  pooling    : pooled Sol finite + positive intercept variance.

Diagnostic note flagged via inline test comment: the ordinal fixture
was redesigned mid-write because the original (round(x1 + 2.5) + 1,
clamped) gave ~zero level-1 observations in training, and BACE
produced level-1 imputations anyway. May relate to the .align_levels
_to_probs fallback path -- worth investigating as a potential source
bug. Switched to balanced quartile-cut fixture to get a clean
regression test, and bumped n=60 with longer chain to absorb noise.

tests/testthat/test-bace_pipeline.R (additions):
The skip_conv = TRUE pipeline test was upgraded from shape-only
(s3 class, non-NULL slots) to NUMERICAL: every imputed dataset's
y_cont must be finite and within 5sd of the column mean (catches
exploding imputations); every y_bin must be a valid factor level;
every pooled Sol entry must be finite. Existing pipeline tests
also got proper numerical assertions on imputed values.

tests/testthat/test-simulate.R (additions, 9 new test_thats):
Validation-path coverage for sim_bace: custom var_names, mismatched
phy... (continued)

2790 of 3670 relevant lines covered (76.02%)

59.97 hits per line

Jobs
ID Job ID Ran Files Coverage
1 25334172491.1 15
76.02
 GitHub Action Run
Source Files on build 25334172491