daniel1noble / BACE / 25338302573
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: 91.161%. Remained the same
25338302573

push

github

daniel1noble 
Per-type accuracy benchmark on simulated data with known truth

dev/sim_accuracy_benchmark.R: per-type sweep across all 5 BACE
response types {gaussian, poisson, binary, K=3 categorical,
K=4 ordinal}. For each, simulates 60 species / 120 cases via
sim_bace() with phylo_signal=0.9, masks 20% of the response
MCAR, runs bace(), and compares imputed-vs-truth using the same
type-aware metrics (NRMSE / balanced_accuracy / mae_level) as
the empirical benchmark engine. Crucially also reports a
column-mean / modal-class baseline so we can show BACE is
actually using the phylogenetic structure.

Single-replicate sweep (~16s on a single core) results:
  gaussian       BACE NRMSE 0.415 vs baseline 0.783 (better)
  poisson        BACE NRMSE 1.193 vs baseline 1.151 (worse)
  binary         BACE bal_acc 0.778 vs baseline 0.500 (better)
  categorical_K3 BACE bal_acc 0.518 vs baseline 0.333 (better)
  ordinal_K4     BACE mae_level 0.417 vs baseline 1.000 (better)

Poisson result is flagged: with only 24 hidden cells per
replicate, the noise floor is large. Manuscript claims should
come from a multi-replicate version (20+ reps per type).
Poisson NRMSE may also be unfair to BACE's posterior-predictive
sampling, which correctly propagates Poisson sampling variance;
consider rank-correlation or coverage95 as alternative headline
metrics for that type.

Env-var BACE_SIM_TYPE=<label> runs a single type for debugging.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

3352 of 3677 relevant lines covered (91.16%)

87.76 hits per line

Jobs
ID Job ID Ran Files Coverage
1 25338302573.1 15
91.16
 GitHub Action Run
Source Files on build 25338302573