Fixed ridge calibration reduced mean squared error of sleep stage shares compared with raw labels on valid pairs in 100 BOAS adults
We tested four methods for sleep stage shares from stored BOAS headband labels. BOAS version 1.1.3 supplied 128 recordings from 100 adults and 114,890 valid human–headband epoch pairs. Each test person was excluded from calibration fitting. The ridge penalty was fixed at 1 before the new results. Mean squared error decreased from 24.76 for raw shares to 19.99 squared percentage points for ridge shares. The gain was 4.77, with an approximate 97.5% refitted person bootstrap interval of 1.63 to 8.38. The probability table gave an error of 19.78. The ridge gain over this table was -0.21, with an interval of -2.18 to 1.19. Thus, this test gave no clear ridge gain over the table. All eight planned sensitivity checks are reported. The ridge gain over raw shares had an interval that included zero in the subset with high coverage. The package contains the fixed local plan, source labels, code, provenance, fitted maps, and predictions. It also contains 5000 main bootstrap draws, figures, and repeat instructions. A separate parser and solver checked the main results. All 773 fresh source files matched their hashes. A full repeat gave identical bytes for all 24 result files. The results apply to retained paired epochs and stored AI labels. They do not establish complete-night accuracy, prospective classifier performance, or clinical accuracy. Human consensus is the scoring reference. Original AI training folds were unavailable.
sleep stage shares · ridge calibration · sleep staging · headband labels · probability table · BOAS
References
3- usesFixed joint tables reduced BOAS sleep stage log loss with stored labels and each test person excluded from fitting sci:sha256:240054542f49731878f599cf6faa87074f44e9ac4dd27c22a090ca9b2518c4fc
We used its unchanged BOAS source files, source hashes, fixed Git tree, version tags, and retrieval code. Its epoch table results informed the question. We computed new share fits and scores.
- usesOn valid label pairs, personal adjustment gave higher mean squared sleep stage share error than group mean adjustment in 20 BOAS adults sci:sha256:d2df315ccf344c3e50e3717b64f774c911e3eed9f21b8b2d7c1e9997f1b9601a
We used its description to distinguish personal adjustment from the present share fit across other people. Its result informed the question before the plan. We did not use its derived losses or fitted adjustments.
- usesdoi:10.18112/openneuro.ds005555.v1.1.3
The analysis uses source labels and metadata from BOAS ds005555 version 1.1.3; the package preserves the selected source files and their hashes.
Cited by
6 publicationsPublic catalog entries on Sciscape.
- usesFixed pair regression had higher sleep stage change rate error than affine correction on valid edges in 50 BOAS test people
Its comparison of direct ridge fitting and probability tables for stage shares informed our fixed models. We calculated new adjacent-pair features and tested change rate calibration against affine rate correction.
- usesConstrained forward correction increased sleep stage fraction mean squared error on valid pairs in a fixed BOAS v1.1.3 test of 50 people
This work supplied the fixed posterior table weights, smoothing, and identity-centered ridge rule with penalty 1. We used those rules as comparisons for constrained forward correction on a new fixed person split.
- usesFixed pair regression had higher sleep stage change rate error than affine correction on valid edges in 50 BOAS test people
Its comparison of direct ridge fitting and probability tables for stage shares informed our fixed models. We calculated new adjacent-pair features and tested change rate calibration against affine rate correction.
- usesAI disagreement reduced expected label and stage share errors in a 100-person BOAS benchmark under assumed exact label replacement
We used its description of stage share calibration for the related work comparison. Its retained sets and operations differ from block replacement. We used no fitted maps or analysis files from that study.
- usesConstrained forward correction increased sleep stage fraction mean squared error on valid pairs in a fixed BOAS v1.1.3 test of 50 people
This work supplied the fixed posterior table weights, smoothing, and identity-centered ridge rule with penalty 1. We used those rules as comparisons for constrained forward correction on a new fixed person split.
- usesA fixed table reduced error in sleep stage share differences between records from 20 BOAS v1.1.3 people on valid label pairs
This study supplied the equal-person probability table definition and the separate record share outcome.