Duration shuffles within nights retained 77% of forecast loss gain in 50 BOAS test people under fixed rules
We tested next-stage forecasts of stored human sleep labels in BOAS version 1.1.3. Fifty people trained fixed probability tables. Fifty other people tested them. The baseline used current stage, recording hour, and preceding different stage. A duration table reduced log loss from 0.31951 to 0.30779 nats per eligible edge. This was a 3.67% reduction. Forty-eight test people improved. An added check shuffled duration bins within nights and predictor cells. Its 200 draws retained 77.33% of the gain on average. A broader shuffle across people and nights increased loss. Thus, the full gain does not isolate the local duration association. The planned refitted bootstrap was poorly centered. We report this limit and a separate fixed-table estimate. The package preserves all 389 source files, hashes, plans, code, 58,386 test predictions, and all result draws. Independent calculations passed. Fresh retrieval and full reruns reproduced 40 numeric files byte for byte. The result applies to these stored consensus labels and fixed rules. It does not establish a neural mechanism.
sleep staging · sleep stage transitions · duration · forecasting · log loss · bootstrap · shuffle diagnostic · BOAS · OpenNeuro
References
2- usesOn common valid epochs from 100 BOAS people, stored AI agreement exceeds headband agreement with human stages sci:sha256:4d7acb36affb5cfe09cfd13e715d5642aa7b2138b6fa15aff94d784cb9f5d795
We used its archived BOAS 1.1.3 source files and pinned source hashes. We selected PSG labels and the person map, then retrieved all 389 selected files again. Its agreement scores did not enter our forecasts.
- usesFixed joint tables reduced BOAS sleep stage log loss with stored labels and each test person excluded from fitting sci:sha256:240054542f49731878f599cf6faa87074f44e9ac4dd27c22a090ca9b2518c4fc
Its fixed label tables supplied a precedent for equal person weight and 1000 effective training counts per person. We used new history tables and a separate 50/50 person split to test next-stage forecasts.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesThe NREM clock gain remained uncertain for five-minute REM forecasts in 50 BOAS test people with fixed logistic models
Its stage duration forecast report informed the current stage controls, person weights, and refit uncertainty check. This study tests a different REM outcome and compares two history clocks. We copied no code or results from that report.
- usesA fixed BOAS duration rule reduced forecast loss after calibration in 31 MRI participants
This study supplied the upstream no-clock BOAS duration rule and source count method. We preserve its unchanged code and fixed source split.
- usesA fixed BOAS duration rule reduced stage change log loss in 31 MRI participants with observed bout starts
We used its preserved BOAS source labels, fixed 50-person training split, original fit code, and no-clock duration rule. We reproduced its source test losses and tested the fixed rule on MRI sleep labels from ds003768.
- usesStored headband label duration reduced 30-second human stage change forecast loss under fixed models in 50 BOAS v1.1.3 test people with both bout starts observed
Its human duration forecast results supplied prior evidence. This study tests headband history and a binary human change target.