Module predict 

Library-side survival prediction pipeline.

Extracts the hazard/survival/cumulative-hazard math from the CLI’s run_predict_survival so that both the CLI and the Python FFI can share a single entry point. The CLI retains ownership of progress bars, CSV writing, and uncertainty bounds; everything else (design build, baseline + time basis evaluation, link/time wiggles, and hazard/survival conversion) flows through predict_survival.

Structs

CompetingRisksPredictResult

Joint cause-specific competing-risks prediction result.

KaplanMeier

Right-continuous Kaplan–Meier survival estimator Ŝ(t) = ∏_{t_j ≤ t}(1 − d_j/n_j).

SurvivalPredictRequest

Inputs to the unified survival predict pipeline.

SurvivalPredictResult

Result of predict_survival.

SurvivalTimeColumns

Resolved survival entry/exit column indices for a saved survival model.

Enums

SurvivalPredictError

Typed errors emitted by the survival prediction pipeline.

Functions

apply_inverse_link_state_to_fit_result

build_saved_survival_marginal_slope_predictor

Build the saved survival marginal-slope predictor along with the matching PredictInput and a UnifiedFitResult repackaged into the layout BernoulliMarginalSlopePredictor::from_unified expects.

concat_array1_refs

Concatenate referenced 1-D arrays into a single owned Array1<f64>.

fit_result_from_saved_model_for_prediction

Canonical saved fit result for prediction.

harrell_concordance

Harrell’s concordance index (C-index) of a survival risk score against held-out outcomes. A larger risk[i] must predict a SHORTER survival time (higher hazard). Over every orderable pair — pairs whose earlier observed time is a genuine event, so the failure ordering is observed — a pair is concordant when the earlier-failing subject carries the larger risk; equal risks score half credit. C = (concordant + 0.5·tied) / comparable. C = 0.5 is random ranking, C = 1.0 a perfect ordering.

integrated_ipcw_brier_score

Integrated IPCW Brier score (IBS) — the time-integrated ipcw_brier_score, matching scikit-survival’s integrated_brier_score and pec’s integrated prediction-error curve.

ipcw_brier_score

IPCW (inverse-probability-of-censoring-weighted) Brier score of a predicted survival probability at a fixed horizon tau against held-out outcomes — the Graf et al. (1999) estimator used by scikit-survival brier_score, pec, and survival::brier.

predict_competing_risks_survival

predict_survival

Run the survival prediction pipeline.

require_saved_survival_likelihood_mode

Extract the saved survival likelihood mode from the model payload.

resolve_saved_survival_time_columns

Resolve saved survival entry/exit column names against the runtime col_map, treating an absent survival_entry as the right-censored shorthand (entry times synthesized as zero downstream).

resolve_survival_inverse_link_from_saved

Resolve the saved survival inverse-link from saved link metadata and fitted state.

resolve_termspec_for_prediction

Resolve the covariate TermCollectionSpec for prediction, remapping saved training-column indices onto the runtime dataset’s layout.

saved_baseline_timewiggle_components

Rebuild the saved baseline-timewiggle entry/exit/derivative design blocks from the saved runtime metadata. Returns None when the saved model has no baseline-timewiggle.

saved_survival_location_scale_fit_result

Resolve the saved survival location-scale fit result.

saved_survival_runtime_baseline_config

Baseline config persisted by the saved survival model.