Parametric

Lifetime distributions with a fixed functional form, fitted by estimating a few parameters. Each distribution is exported as a ready-made instance (surpyval.Weibull, surpyval.LogNormal, …). Its fit accepts any mix of observed, censored and truncated data and returns a Parametric model; its from_params builds the same model from known parameters. The distribution’s own functions (sf, ff, df, hf, Hf, qf, mean, moment, random, …) can also be called directly with the parameters as extra arguments, for example Weibull.sf(x, alpha, beta).

fit supports five estimation methods (how='MLE', 'MPP', 'MOM', 'MSE', 'MPS'), fixed parameters (fixed), an offset (offset=True), a limited failure population (lfp=True) and zero inflation (zi=True); fit_best() fits every candidate distribution and keeps the best (see Comparison Tests and Validation Metrics). The theory is in Parametric Estimation and worked examples are in Parametric SurPyval Modelling.

Parametric Class

The fitted model every distribution’s fit and from_params returns.

Distribution Classes

Continuous lifetime distributions.

Discrete Distribution Classes

Discrete lifetimes on the positive integers, for cycle- or demand-counted data. All accept the same censoring and truncation formats as the continuous distributions.

Special Distributions

Per-demand and degenerate models with no (or a fixed) time dimension, for composing into mixtures, competing risks and demand studies.

Flexible Parametric (Royston-Parmar)

A spline model for data whose hazard no standard distribution fits.

Custom Distributions

Define a new distribution from its cumulative hazard function alone.

Zero-Failure Analysis (Weibayes)

The bound on a Weibull scale of known shape when there are too few failures to fit one, including none.

Mixture Modelling

A population made of several sub-populations, each with its own distribution of the same family.