Compress a posterior draws matrix

Fits a compact density approximation to a matrix of MCMC draws and returns an object of class posterior_compressed. The compressed object can be sampled from via sample_posterior() and evaluated via density_posterior().

compress_posterior(
  draws,
  method = c("mclust", "mvdens_gmm", "mvdens_kde"),
  variables = NULL,
  n_components = 3L,
  model_name = NULL,
  verbose = FALSE,
  ...
)

Arguments

draws

An object containing posterior draws. Accepted:

• a numeric matrix (rows = draws, cols = parameters);

• a data.frame of draws (one column per parameter);

• any object the posterior package can convert with posterior::as_draws_matrix() (draws_matrix, draws_array, ...);

• a 3-D array of shape iter x chain x var.

method

One of 'mclust', 'mvdens_gmm', 'mvdens_kde'. See compression_methods().

variables

Optional character vector of parameter names to keep. If NULL (default) all columns are used.

n_components

Integer number of mixture components (used by "mclust" and "mvdens_gmm"). Default 3.

model_name

mclust covariance structure (e.g. "VVV", "EEE", or a vector of allowed model names). Ignored by other methods. When NULL (default) poco auto-selects a sensible set: the spherical and diagonal models c("EII", "VII", "EEI", "EVI", "VEI", "VVI") are used when nrow(draws) <= ncol(draws) so covariances remain identifiable, otherwise mclust's full default set is used and BIC picks the best.

verbose

Logical; print backend progress. Default FALSE.

...

Additional arguments forwarded to the backend (e.g. mclust::Mclust()).

Value

An S3 list of class c("posterior_compressed_<method>", "posterior_compressed", "list") containing the parameters of the fitted approximation.

See also

compress_fit(), compress_brmsfit(), compress_sccomp(), sample_posterior(), density_posterior().

Examples

set.seed(1)
draws <- matrix(rnorm(2000 * 3), ncol = 3,
                dimnames = list(NULL, c("alpha", "beta", "sigma")))
comp <- compress_posterior(draws, method = "mclust", n_components = 2)
#> ℹ mclust: trying all 14 covariance models c(EII, VII, EEI, VEI, EVI, VVI, EEE, VEE, EVE, VVE, EEV, VEV, EVV, VVV) and picking the best by BIC (n = 2000, d = 3).
#> ℹ mclust: selected model 'EII' with G = 2 (BIC = -17,298.09) out of 14 candidate models: EII, VII, EEI, VEI, EVI, VVI, EEE, VEE, EVE, VVE, EEV, VEV, EVV, VVV.
comp
#> <posterior_compressed: mclust >
#>  parameters: 3
#>  components: 2
#>  original draws: 2000
#>  mclust model: EII
#>  BIC: -17298.09
new_samples <- sample_posterior(comp, n_draws = 500)
dim(new_samples)
#> [1] 500   3