This vignette walks through the full workflow for compressing a
brms::brmsfit, then reconstructing it for use with the
standard brms post-processing functions
(predict(), posterior_predict(),
pp_check(), …). It requires brms +
cmdstanr + a working CmdStan install.
The chunks below that fit Stan models are gated with
instantiate::stan_cmdstan_exists(), so the vignette still
builds on machines that do not have CmdStan installed (e.g. CI runners).
Install CmdStan once with:
poco::check_and_install_cmdstanr()
cmdstanr::check_cmdstan_toolchain(fix = TRUE)
cmdstanr::install_cmdstan()Fit a brms model
The model must use the cmdstanr backend so that
compress_brmsfit() can reach the CSV draws.
library(brms)
#> Loading required package: Rcpp
#> Loading 'brms' package (version 2.23.0). Useful instructions
#> can be found by typing help('brms'). A more detailed introduction
#> to the package is available through vignette('brms_overview').
#>
#> Attaching package: 'brms'
#> The following object is masked from 'package:stats':
#>
#> ar
library(dplyr)
#>
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#>
#> filter, lag
#> The following objects are masked from 'package:base':
#>
#> intersect, setdiff, setequal, union
set.seed(1)
n_groups <- 20; n_per <- 15
dat <- expand.grid(group = 1:n_groups, rep = 1:n_per)
dat$x <- rnorm(nrow(dat))
re <- rnorm(n_groups, 0, 0.5)
dat$y <- 2 + 1.5 * dat$x + re[dat$group] + rnorm(nrow(dat), 0, 0.8)
fit <- brm(
y ~ x + (1 | group),
data = dat,
backend = "cmdstanr",
chains = 4,
iter = 2000,
warmup = 1000,
refresh = 0
)
#> Start sampling
#> Running MCMC with 4 sequential chains...
#>
#> Chain 1 finished in 0.4 seconds.
#> Chain 2 finished in 0.4 seconds.
#> Chain 3 finished in 0.4 seconds.
#> Chain 4 finished in 0.4 seconds.
#>
#> All 4 chains finished successfully.
#> Mean chain execution time: 0.4 seconds.
#> Total execution time: 2.0 seconds.
#> Loading required namespace: rstanCompress and save
A single call returns both the compressed posterior and the brms fit structure. Save them separately so the structure can be reused with any future updates to the compressed posterior.
result <- compress_brmsfit(
fit,
method = "mclust",
n_components = 5
)
#> 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 = 4000, d = 27).
#> ℹ mclust: selected model 'VEV' with G = 5 (BIC = 205,090.9) out of 14 candidate models: EII, VII, EEI, VEI, EVI, VVI, EEE, VEE, EVE, VVE, EEV, VEV, EVV, VVV.result$compressed carries the GMM parameters;
result$structure is the original brmsfit,
which we keep so that downstream methods can be re-attached.
Reconstruct and use the model
fit_recon <- reconstruct_brmsfit(result)
new_data <- data.frame(x = seq(-2, 2, length.out = 20),
group = rep(1:5, each = 4))
p <- predict(fit_recon, newdata = new_data)
pp <- posterior_predict(fit_recon, newdata = new_data)Compare pairwise posterior geometry
Use bayesplot::mcmc_pairs() to compare the original and
reconstructed draw geometry on the same subset of parameters.
library(bayesplot)
#> This is bayesplot version 1.15.0
#> - Online documentation and vignettes at mc-stan.org/bayesplot
#> - bayesplot theme set to bayesplot::theme_default()
#> * Does _not_ affect other ggplot2 plots
#> * See ?bayesplot_theme_set for details on theme setting
#>
#> Attaching package: 'bayesplot'
#> The following object is masked from 'package:brms':
#>
#> rhat
pair_pars <- c("b_Intercept", "b_x", "sd_group__Intercept", "sigma")
draws_orig <- posterior::as_draws_array(fit, variable = pair_pars)
draws_recon <- posterior::as_draws_array(fit_recon, variable = pair_pars)
bayesplot::mcmc_pairs(draws_orig, pars = pair_pars)
bayesplot::mcmc_pairs(draws_recon, pars = pair_pars)
MCMC diagnostics (Rhat, ESS) are not preserved
by the regenerated draws.
Other methods
compress_brmsfit() shares the same method
argument as compress_posterior(). To use a different
backend, e.g. mvdens GMM:
result <- compress_brmsfit(
fit,
method = "mvdens_gmm",
n_components = 5
)(Requires the suggested mvdens package -
remotes::install_github("NKI-CCB/mvdens").)
Session info
sessionInfo()
#> R version 4.6.0 (2026-04-24)
#> Platform: x86_64-pc-linux-gnu
#> Running under: Ubuntu 24.04.4 LTS
#>
#> Matrix products: default
#> BLAS: /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
#> LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so; LAPACK version 3.12.0
#>
#> locale:
#> [1] LC_CTYPE=C.UTF-8 LC_NUMERIC=C LC_TIME=C.UTF-8
#> [4] LC_COLLATE=C.UTF-8 LC_MONETARY=C.UTF-8 LC_MESSAGES=C.UTF-8
#> [7] LC_PAPER=C.UTF-8 LC_NAME=C LC_ADDRESS=C
#> [10] LC_TELEPHONE=C LC_MEASUREMENT=C.UTF-8 LC_IDENTIFICATION=C
#>
#> time zone: UTC
#> tzcode source: system (glibc)
#>
#> attached base packages:
#> [1] stats graphics grDevices utils datasets methods base
#>
#> other attached packages:
#> [1] bayesplot_1.15.0 dplyr_1.2.1 brms_2.23.0 Rcpp_1.1.1-1.1
#> [5] poco_0.2.0
#>
#> loaded via a namespace (and not attached):
#> [1] gtable_0.3.6 tensorA_0.36.2.1 QuickJSR_1.9.2
#> [4] xfun_0.57 bslib_0.10.0 ggplot2_4.0.3
#> [7] htmlwidgets_1.6.4 processx_3.9.0 inline_0.3.21
#> [10] lattice_0.22-9 callr_3.7.6 vctrs_0.7.3
#> [13] tools_4.6.0 ps_1.9.3 generics_0.1.4
#> [16] stats4_4.6.0 parallel_4.6.0 tibble_3.3.1
#> [19] cmdstanr_0.9.0 pkgconfig_2.0.3 Matrix_1.7-5
#> [22] data.table_1.18.4 checkmate_2.3.4 RColorBrewer_1.1-3
#> [25] S7_0.2.2 desc_1.4.3 RcppParallel_5.1.11-2
#> [28] distributional_0.7.0 lifecycle_1.0.5 compiler_4.6.0
#> [31] farver_2.1.2 stringr_1.6.0 textshaping_1.0.5
#> [34] Brobdingnag_1.2-9 codetools_0.2-20 htmltools_0.5.9
#> [37] mvdens_1.1.0 sass_0.4.10 yaml_2.3.12
#> [40] pillar_1.11.1 pkgdown_2.2.0 jquerylib_0.1.4
#> [43] cachem_1.1.0 StanHeaders_2.32.10 bridgesampling_1.2-1
#> [46] abind_1.4-8 mclust_6.1.2 nlme_3.1-169
#> [49] rstan_2.32.7 posterior_1.7.0 tidyselect_1.2.1
#> [52] digest_0.6.39 mvtnorm_1.3-7 stringi_1.8.7
#> [55] reshape2_1.4.5 labeling_0.4.3 fastmap_1.2.0
#> [58] grid_4.6.0 cli_3.6.6 magrittr_2.0.5
#> [61] loo_2.9.0 pkgbuild_1.4.8 withr_3.0.2
#> [64] scales_1.4.0 backports_1.5.1 rmarkdown_2.31
#> [67] matrixStats_1.5.0 otel_0.2.0 gridExtra_2.3
#> [70] instantiate_0.2.3 ragg_1.5.2 coda_0.19-4.1
#> [73] evaluate_1.0.5 knitr_1.51 rstantools_2.6.0
#> [76] rlang_1.2.0 glue_1.8.1 jsonlite_2.0.0
#> [79] plyr_1.8.9 R6_2.6.1 systemfonts_1.3.2
#> [82] fs_2.1.0