Diagnostics attached to fitted values or response samples describe the
coefficient chains before transformation or response simulation. Response
noise can conceal poor mixing of coefficient chains. Use as_mcmc_draws()
to examine particular derived quantities as well.
Usage
sampling_diagnostics(x)
# S3 method for class 'sampling_diagnostics'
print(x, ...)Value
A sampling_diagnostics list with chains (acceptance rates and
elapsed seconds), coefficients (posterior summaries, or NULL if not
requested), and draws (a coefficient matrix with chain identities, or
NULL if unavailable). Coefficient matrices provide their own draws;
other sampling outputs need diagnostics = TRUE to retain them.
Likelihood matrices from log_lik() also require diagnostics = TRUE
for coefficient plots; use as_mcmc_draws() for their log-density draws.
Details
Coefficient draws are retained when sampling with diagnostics = TRUE,
allowing draw.sampling_diagnostics() to plot the original coefficient
chains even when x contains response samples, fitted values, smooths or
derivatives. These plots do not describe the transformed quantities.
Numerical summaries require posterior, but plotting does not.