Development build. This page describes main, not a released package. The latest published Lodestar.Survival is 0.2.0 — read its documentation.

Home › Survival › Survival estimators

AftOptions

What an AcceleratedFailureTime fit may be told.

public sealed record AftOptions

Properties — ConfidenceLevel is the two-sided level the intervals are reported at; 0.95 by default. MaximumIterations is how many Newton steps the fit may take; 100 by default. FitIntercept is lifelines' fit_intercept: whether the primary parameter's linear predictor carries an intercept, true by default. Ancillary is lifelines' ancillary=True: whether the ancillary parameter is modelled by the same covariates, false by default, an intercept alone. Penalizer is lifelines' penalizer at l1_ratio=0, a ridge penalty's weight, zero by default. Robust asks for the Huber sandwich standard errors, lifelines' robust=True, false by default.

Exceptions — ArgumentOutOfRangeException when ConfidenceLevel does not lie strictly inside (0, 1), when MaximumIterations is below one, or when Penalizer is negative, infinite or NaN. Each is thrown where the setting is set, not where the fit reads it.

Example — the dose also moving the Weibull's shape: one more coefficient, on rho_.

using Lodestar.Survival;

double[] dose = [1.0, 2.0, 0.5, 3.0, 0.0, 1.5, 0.0, 2.5, 1.0, 0.5];
double[] months = [5, 8, 12, 3, 15, 9, 20, 6, 11, 14];
bool[] died = [true, true, false, true, true, true, false, true, true, false];

AftSummary fit = AcceleratedFailureTime.Fit(AftModel.Weibull, dose, months, died, 1,
    new AftOptions { Ancillary = true });

int coefficients = fit.Coefficients.Count;                  // => 4
string third = fit.ParameterNames[2];                       // => rho_
int thirdColumn = fit.CovariateIndices[2];                  // => 0
double onShape = Math.Round(fit.Coefficients[2], 6);        // => 0.173917

Remarks — Ancillary works for all three models. lifelines' LogLogisticAFTFitter constructor does not pass its model_ancillary on, so the log-logistic's beta_ is an intercept there whatever is asked; here the setting is read for every AftModel.

The penalty is the ridge part of lifelines' elastic net, penalizer · Σ β² / 2 on the coefficients of the covariates scaled by their sample deviations, added to the mean negative log-likelihood; the L1 part is not written, for the reason docs/equivalence.md gives. An intercept is left unpenalised only when its block holds covariates, as lifelines leaves it: by default the ancillary block is an intercept alone, so its rho_ intercept is penalised and shrinks towards zero, a Weibull shape towards one. LogLikelihood is then the penalised one, as lifelines' log_likelihood_ is.

Robust is lifelines' Huber sandwich: the inverse information either side of the summed outer products of each subject's score, its penalty gradient included. It changes the standard errors and everything built on them, never the coefficients.

Applies to — net10.0, netstandard2.0.

See also — AcceleratedFailureTime.Fit, AftSummary.