Reference / binference.validation.sbc
binference.validation.sbc¶
- binference.validation.sbc(simulate, fit, n_simulations=100, *, n_draws=99, log_likelihood=True, rng=None, progress=False)[source]¶
Simulation-based calibration.
- Parameters:
- simulatecallable
simulate(rng) -> (likelihoods, true_value)drawing a population from the prior and a catalogue from it (seebinference.simulate.prior_predictive()). The true value is the fractions, or the rates for rate models.- fitcallable
fit(likelihoods, rng) -> PopulationResultrunning the inference to be validated (with the same prior as the simulator).- n_simulationsint
Number of simulated data sets.
- n_drawsint
Posterior draws ranked per fit, thinned evenly from its draws. They should be roughly independent, i.e. fewer than the ESS: a warning names how many fits have a smallest bulk ESS below it, since autocorrelated draws make the rank histograms look over-dispersed.
- log_likelihoodbool
Also rank the joint log-likelihood of the true population.
- rngint, Generator or None
Seed or generator of the simulations and fits.
- progressbool
Show a progress bar over the simulations.