NEWS

survlab 0.2.0 (2026-10-22)

impute_nondetect() gains a control argument accepting a survival::survreg.control() object. This allows users to tune the underlying survreg fitting algorithm - most notably maxiter - for datasets where the default iteration limit causes convergence warnings.
impute_nondetect() gains a zero_value argument (default 1e-5): exact zeros are replaced by it when fitting the log-normal, Weibull, exponential and log-logistic distributions, which previously failed silently on data containing zeros. The zeros are kept in the final values.
impute_nondetect() now records the AIC and convergence of every distribution (attribute fits) and selects the best model among the converged fits only. If none converged, the best non-converged fit is used, a warning is raised, and attribute best_converged is FALSE. Convergence warnings from the individual fits are no longer passed on to the caller.
Fixed: with min_value = 0, logistic imputations were not truncated at 0, and draws below 0 were set to exactly 0. They are now drawn from the logistic truncated between min_value and the detection limit.