You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Debug create_adjusted_data() for dcurves example #127
Rechecked this against current main before changing implementation.
The discrepancy is no longer reproducible. On the dcurvesdf_surv example at 1.5 years, current rtichoke matches dcurves for threshold-specific TP rate, FP rate, and net benefit at representative cutoffs (0, .01, .05, .10, .20, .50). The reported prevalence is ~0.22, consistent with 165 / 750 (and 585 estimated negatives). An exact-boundary regression also confirms the prediction >= threshold convention.
The key subsequent change is that time-dependent performance now uses risk_set_scope=["pooled_by_cutoff"], so KM/AJ quantities are estimated on the whole predicted-positive/predicted-negative groups at each cutoff instead of being reconstructed from small probability bins.
Regression coverage is in PR #327. No implementation change is warranted.
Activity
adjusted_as_composite#120uriahf commentedon Aug 20, 2026
Rechecked this against current
mainbefore changing implementation.The discrepancy is no longer reproducible. On the
dcurvesdf_survexample at 1.5 years, current rtichoke matchesdcurvesfor threshold-specific TP rate, FP rate, and net benefit at representative cutoffs (0, .01, .05, .10, .20, .50). The reported prevalence is ~0.22, consistent with 165 / 750 (and 585 estimated negatives). An exact-boundary regression also confirms theprediction >= thresholdconvention.The key subsequent change is that time-dependent performance now uses
risk_set_scope=["pooled_by_cutoff"], so KM/AJ quantities are estimated on the whole predicted-positive/predicted-negative groups at each cutoff instead of being reconstructed from small probability bins.Regression coverage is in PR #327. No implementation change is warranted.
test: lock dcurves parity for issue #127