Commit bb0dcee
committed
[ML] Relax Bayesian optimisation mean improvement threshold for debug builds
The testMaximumExpectedImprovement test requires BO mean improvement to
exceed 1.5x random search, but with the reduced iteration count in debug
builds (15 vs 30) BO doesn't accumulate enough advantage to consistently
hit that bar. This has been failing deterministically on macOS and Windows
in every nightly debug build.
Introduce a MEAN_IMPROVEMENT_MULTIPLIER constant (1.5 for release, 1.2
for debug) alongside the existing debug-mode relaxations for NUM_TRIALS,
NUM_BO_ITERATIONS, and WIN_RATE_THRESHOLD.
Made-with: Cursor1 parent 65432f4 commit bb0dcee
1 file changed
+7
-2
lines changed| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
286 | 286 | | |
287 | 287 | | |
288 | 288 | | |
| 289 | + | |
289 | 290 | | |
290 | 291 | | |
291 | 292 | | |
292 | | - | |
| 293 | + | |
| 294 | + | |
| 295 | + | |
293 | 296 | | |
294 | 297 | | |
295 | 298 | | |
| 299 | + | |
296 | 300 | | |
297 | 301 | | |
298 | 302 | | |
| |||
384 | 388 | | |
385 | 389 | | |
386 | 390 | | |
387 | | - | |
| 391 | + | |
| 392 | + | |
388 | 393 | | |
389 | 394 | | |
390 | 395 | | |
| |||
0 commit comments