mcmcChecks: MCMC Checks

In imt: Impact Measurement Toolkit

MCMC Checks

Description

Checks convergence, mixing, effective sample size, and divergent transitions

Methods

$new(fit, pars)

Runs diagnostics on the supplied stanfit object, restricted to parameters identified by the character vector pars.
Tests include:
Share of specified parameters with an Rhat less than 1.1. If any have an Rhat > 1.1, everything_looks_fine is set to FALSE.
Share of specified parameters with an n_eff at least 0.1% of the total number of posterior draws. If any have n_eff < 0.001 * N, everything_looks_fine is set to FALSE.
Share of specified parameters with an n_eff of at least 100. If any have n_eff < 100, everything_looks_fine is set to FALSE.
Number of divergent transitions during posterior sampling. If there are any whatsoever, everything_looks_fine is set to FALSE.
Share of posterior iterations where the sampler reached the maximum treedepth. If more than 25\ everything_looks_fine is set to FALSE.

Active bindings

everything_looks_fine

logical indicating whether all MCMC tests passed.

diagnostics

list of the outcome of each MCMC test

warnings

list of the warning messages from failed MCMC tests

Methods

Public methods

• mcmcChecks$new()

• mcmcChecks$clone()

---

Method new()

Initialize a new mcmcChecks object and run diagnostics

Usage

mcmcChecks$new(fit, pars)

Arguments

fit

A stanfit object to check

pars

A character vector of parameter names to check

---

Method clone()

The objects of this class are cloneable with this method.

Usage

mcmcChecks$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

imt documentation built on Sept. 11, 2024, 5:41 p.m.