Converts the retained MCMC samples of every chain into a
posterior::draws_array (dimensions iteration \(\times\) chain
\(\times\) variable) of permutation-invariant scalar summaries. This is the
reusable primitive behind convergence_diagnostics: because it
returns a standard draws_array, any function in the posterior
package can be applied to it directly.
Diagnostics are computed on invariant summaries rather than the raw
A/D matrices because strain labels are not identifiable across
iterations or chains, which makes element-wise R-hat / ESS on A or
D uninterpretable.
Usage
as_draws_snp_slice(
results,
pars = c("logpost", "n_strains", "kstar", "ktrunc"),
additional_burnin = 0
)Arguments
- results
A
snp_slice_resultsobject (withstore_mcmc = TRUE).- pars
Character vector of parameters to include. One of
"logpost","n_strains","kstar","ktrunc", which are all scalar per sample, or"coi"(per-host complexity of infection), which is a vector that expands tocoi[1]..coi[N]).- additional_burnin
Number of additional stored samples to discard from the start of each retained chain before assembling the array.
Details
All chains are pooled into one array so that R-hat and ESS are computed for the
whole run. Without early stopping every chain retains the same number of
samples; when gap early-stopping fired, chains can differ in length, in
which case every chain is truncated to the shortest and a warning is issued.
Examples
result <- load_example_results()
draws <- as_draws_snp_slice(result)
posterior::summarise_draws(draws)
#> # A tibble: 4 × 10
#> variable mean median sd mad q5 q95 rhat ess_bulk ess_tail
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 logpost -7.51e4 -75010. 123. 109. -75254. -74903. 2.19 4.01 41.7
#> 2 n_strai… 5.11e1 52 2.16 1.48 48 53 7.59 3.32 3.52
#> 3 kstar 1.13e2 113 0 0 113 113 NA NA NA
#> 4 ktrunc 1.34e2 134 0 0 134 134 NA NA NA