Skip to content

Latest commit

 

History

History
41 lines (36 loc) · 3.06 KB

File metadata and controls

41 lines (36 loc) · 3.06 KB

Gap closures (research workbench)

Shipped in this pass vs intentionally deferred. Framing: research / education only — not a clinical device.

Shipped

Area What
Per-run trust pack trust_pack.json under reports/runs/<id>/; leakage/SHAP/ext-val bind to run_id; promote syncs trust artifacts; ZIP/methods.md accept run_id
External validation job POST /v1/jobs/external-validate + Results UI form
Analysis pack GET /v1/reports/analysis-pack + Analytics KPIs/download
Leakage ICD warnings warnings[] when ICD/diagnosis-style feature names appear
Horizon / teaching tasks horizon_detection_30d, teaching_leaky_contrast + demo CSV
Task-aware health GET /v1/datasets/health?task_id= enforces required columns
Experiment browser GET /v1/runs, GET /v1/runs/{id}, promote; Results UI list / open / compare / promote
Fairness panel GET /v1/reports/fairness, Results “Run fairness” + group table (age bands / TPR·FPR); summary + ZIP
Light HPO POST /v1/jobs/hpo + Train optional light grid; Results best-trial card + trials table (n/a for unavailable metrics); reports/hpo_report.json
Demo dataset catalog Teaching fixtures under data/demo/; Train defaults to data/demo/ehr_data.csv; legacy raw paths resolve as fallbacks
Dataset browse / delete Demo toggle, tabular selection, multi-delete; missing allowed paths return idempotent already_absent
Job UX Cancel (queued/best-effort), recent jobs list on Train, clearer status
Task clarity Task description + required_columns (readmission_30d highlights index_time)
Auth empty states Global 401 banner → Config; Predict already CTAs to Train when no schema
LSTM Not selectable in Train/Config; meta lists under unsupported
Thresholds GET /v1/reports/thresholds + Results table (operating points; optional reports/threshold_operating_points.json)
Trust pack Per-run reports/runs/<id>/trust_pack.json + run detail flags (not /v1/reports/trust-pack)
Docs This note · RESEARCH_WORKFLOW.md · brief website updates (ui-analytics / features)
Analytics export / print Per-chart PNG + print/PDF layout on Analytics (research appendix figures; not clinical reporting)
Research wizard /research guided loop (health → train → trust → leakage → external → export)
Paper curves ROC / PR / calibration points in evaluation_report.json; Analytics Chart.js plots + GET /v1/reports/curves; bootstrap AUC CIs + quality note
Predict session export Predict page → Export session JSON (features + score; research only)
Playwright smoke Optional web/e2e (npm run e2e with UI up)

Deferred

Item Why
Full Optuna / AutoML Out of scope; light grid only
Hard-kill of running sklearn fits Best-effort cancel flag only
LSTM / Transformers path Explicit non-goal (LIMITATIONS.md)
Clinical decision thresholds Research operating points only
Multi-tenant auth / SaaS Local optional API_KEY only