Bootstrap Uncertainty¶

Semiparametric Bootstrap¶

Resamples death counts from their assumed distribution (Poisson or Binomial) with mean equal to the fitted deaths, then refits the model on each sample:

import pystmomo as ps

data = ps.load_ew_male()
fit  = ps.lc().fit(data.deaths, data.exposures,
                   ages=data.ages, years=data.years)

boot = ps.semiparametric_bootstrap(fit, nboot=500, seed=0)
print(boot)
# BootStMoMo(method='semiparametric', nboot=500, n_fits_ok=500)

Residual Bootstrap¶

Resamples deviance residuals and inverts them to obtain bootstrap death counts (Renshaw & Haberman 2008):

boot_r = ps.residual_bootstrap(fit, nboot=500, seed=0)

Confidence Intervals¶

lo, hi = boot.parameter_ci("kt", level=0.95)
lo_bx, hi_bx = boot.parameter_ci("bx", level=0.95)

Available parameters: "ax", "bx", "kt", "b0x", "gc".

Parallel Bootstrap¶

pip install "pystmomo[parallel]"

boot = ps.semiparametric_bootstrap(fit, nboot=500, n_jobs=-1, seed=0)

Accessing Raw Replicate Fits¶

All fitted FitStMoMo objects are stored in boot.fits:

deviances = [f.deviance for f in boot.fits]