Feature reset#6
Open
Suadeo03 wants to merge 19 commits into
Open
Conversation
- Add .gitignore and .dockerignore for Python/ML project - Add CLAUDE.md with project context, constraints, and feature map - Add build_dev_subset.py for stratified dev data sampling - Swap RandomForest → XGBoost in team_code.py and requirements.txt Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Both files contain local dev context that should not be in the shared repo. Added to .gitignore so they won't be re-staged. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…gn, tighter regularization (48 features, LOSO 0.540)
…C 0.5399 unchanged); reorganize EDA scripts)
- Add AgeResidualizer pipeline step (features/age_residuals.py): appends CA_rate_age_residual and EEG_var_REM_Wake_age_residual (feature count 48→50) - Extract build_pipeline() to features/pipeline.py — shared by team_code.py and loso_cv.py, closing the divergence bug where LOSO validated a stale 48-feature model while train_model had already moved to 50 - Add ratchet check system (ratchet_check.py, ratchet_baselines.json) and wire into check_submission_files.py for pre-submission regression gating - Add reg_sweep.py for regularization coefficient sweeps Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- git mv check_submission_files.py, ratchet_check.py, ratchet_baselines.json, reg_sweep.py into tools/; plain mv for untracked loso_cv.py and confusion_at_threshold.py - Fix sys.path.insert in loso_cv.py and reg_sweep.py to walk up two levels (tools/../ = repo root) so from features/... and from helper_code/... resolve - Add _DEFAULT_BASELINES_PATH in ratchet_check.py (script-relative, not CWD-relative) and use it as default for load_baselines(), check_ratchet(), and --baselines-file argparse arg - Same script-relative fix for check_submission_files.py --baselines-file - Anchor /loso_cv.py in .gitignore so tools/loso_cv.py is now tracked All five verification commands (check_submission_files, loso_cv --help, reg_sweep --help, ratchet_check standalone, check_submission_files + ratchet) pass from repo root. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The reconciled loso_cv.py had sys.path.insert at line 35 (before imports), not line 63 where the earlier edit landed. Patch now inserts the correct one-level-up dirname so helper_code and features/ resolve from repo root when the script is invoked from inside tools/. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
features/caisr_enriched.py: - Add extract_markov_transition_features() / N_MARKOV_FEATURES=9 (NF1): transition_entropy, n3_to_wake_rate, rem_to_n3_rate, escalating_rate, + 5-state stationary distribution (Wake/N1/N2/N3/REM). Gated on _MIN_TRANSITIONS=40 (estimability of transition matrix, not duration — see module-level "Gating design decision" comment for why duration was rejected as the gate). NOT wired into extract_caisr_enriched_features. - Add extract_arousal_clustering_features() / N_AROUSAL_CLUSTERING_FEATURES=2 (NF2): cv_inter_arousal_interval + burst_index. Gated on _MIN_AROUSAL_EVENTS=5. NOT wired in. - Document the _MIN_RECORDING_HOURS gating-design decision: confirmed the duration guard described in 2026-07-12 learning_log.md was never implemented (not a lost commit); resolved here by gating on estimability directly rather than reviving a blanket duration cutoff. tools/test_combined_features.py: - New standalone ablation script: tests NF1+NF2 as a single 11-feature block against the shipped 48-feature baseline (NOT one-at-a-time). Collinearity check runs and prints BEFORE the ablation. Supports --rotate-i0002, --C, consistent with loso_cv.py/reg_sweep.py conventions. - Fix: corrected _IDX_WAKE/N1/N2/N3/REM_PCT index constants (were off by 1 — FEATURES.md comment incorrectly said caisr_base[2]=Wake%, but caisr_base[2]=Limb_idx; Wake% starts at caisr_base[3]). Verified against caisr_base.py's actual return array. _IDX_AROUSAL_IDX and _IDX_SPONT_AROUSAL were correct and unchanged. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
FeatureDropper (features/feature_selection.py) inserted into pipeline after AgeResidualizer; removes 11 features whose coefficients changed sign across LOSO folds — measured evidence of site-specific overfitting. LOSO result: +8.2% relative reward, AUROC 0.6459→0.6711, all three folds improved, cross- fold spread shrank 0.1264→0.1145. Threshold re-tuned 0.08→0.10 for new probability distribution. features/subtype_weighting.py (Entry 7 MCI weighting, validated separately) added to repo but not wired into this entry. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Entry 8: drop sign-flipping features (Stage 1), threshold 0.10
Author
|
Merged feature add updates |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.