Ggm mixed

.Rbuildignore

@@ -16,6 +16,11 @@
 
 # GitHub / CI
 ^\.github$
+^CONTRIBUTING\.md$
+
+# Hidden config files (development only)
+^\.lintr$
+^\.editorconfig$
 
 # R CMD build artifacts
 ^doc$

.github/copilot-instructions.md

@@ -9,13 +9,7 @@ Rules for AI agents (Copilot, Claude, etc.) working on this codebase.
   be captured, use `<-`. With `=` R treats it as a named argument.
   Example: `expect_message(result <- foo(), "pattern")`.
 - No space between `if`/`for`/`while` and `(`: write `if(`, not `if (`.
-- Enforced by `inst/styler/bgms_style.R`. Run before committing:
-  ```r
-  source("inst/styler/bgms_style.R")
-  styler::style_pkg(style = bgms_style)
-  ```
-  After running, check test files for `expect_*(result = ...)` and
-  revert those to `result <- ...`.
+- Enforced by `inst/styler/bgms_style.R` (see Pre-commit checks).
 
 ## Exported R functions (Tier 1)
 
@@ -80,6 +74,40 @@ Rules for AI agents (Copilot, Claude, etc.) working on this codebase.
   the same commit.
 - When adding a new exported function, add it to `_pkgdown.yml`.
 
+## Pre-commit checks
+
+Before every commit that touches R code, run these checks and fix
+any issues they report:
+
+1. **Style** — enforce the project code style:
+   ```r
+   source("inst/styler/bgms_style.R")
+   styler::style_pkg(style = bgms_style)
+   ```
+   After running, check test files for `expect_*(result = ...)` and
+   revert those to `result <- ...`.
+
+2. **Lint** — catch issues the CI lint workflow will flag:
+   ```r
+   lintr::lint_package()
+   ```
+   The `.lintr` config enables `T_and_F_symbol_linter()` and
+   `seq_linter()` (among defaults). Fix all findings before
+   committing.
+
+3. **Roxygen** — regenerate Rd files if any roxygen comment changed:
+   ```r
+   roxygen2::roxygenise()
+   ```
+   Stage the regenerated `man/*.Rd` files in the same commit.
+
+4. **R CMD check** — if the change is non-trivial, run:
+   ```r
+   rcmdcheck::rcmdcheck(args = c("--no-manual", "--as-cran"))
+   ```
+   CI uses `error-on = "warning"`, so warnings are treated as
+   failures.
+
 ## Do not
 
 - Add `@keywords internal` to exported functions.

.github/workflows/nightly-validation.yaml

@@ -0,0 +1,29 @@
+name: nightly-validation
+on:
+  schedule:
+    - cron: '0 3 * * 1,4'  # Monday and Thursday at 3 AM UTC
+  workflow_dispatch: {}
+
+jobs:
+  validate:
+    runs-on: ubuntu-latest
+    timeout-minutes: 120
+    env:
+      GITHUB_PAT: ${{ secrets.GITHUB_TOKEN }}
+      R_KEEP_PKG_SOURCE: yes
+      BGMS_RUN_SLOW_TESTS: true
+
+    steps:
+      - uses: actions/checkout@v5
+
+      - uses: r-lib/actions/setup-r@v2
+        with:
+          use-public-rspm: true
+
+      - uses: r-lib/actions/setup-r-dependencies@v2
+        with:
+          extra-packages: any::devtools
+
+      - name: Run full test suite (including slow tests)
+        run: Rscript -e 'devtools::test()'
+

.github/workflows/weekly-compliance.yaml

@@ -0,0 +1,35 @@
+name: weekly-compliance
+on:
+  schedule:
+    - cron: '0 5 * * 0'  # Sunday at 5 AM UTC
+  workflow_dispatch: {}
+
+jobs:
+  compliance:
+    runs-on: ubuntu-latest
+    timeout-minutes: 120
+    env:
+      GITHUB_PAT: ${{ secrets.GITHUB_TOKEN }}
+      R_KEEP_PKG_SOURCE: yes
+
+    steps:
+      - uses: actions/checkout@v5
+
+      - uses: r-lib/actions/setup-r@v2
+        with:
+          use-public-rspm: true
+
+      - uses: r-lib/actions/setup-r-dependencies@v2
+
+      - name: Install mixedGM (needed for cross-package tests)
+        run: Rscript -e 'remotes::install_github("MaartenMarsman/mixedGM")'
+
+      - name: Generate compliance fixtures (if missing)
+        run: |
+          if [ ! -f tests/compliance/fixtures/manifest.rds ]; then
+            Rscript tests/compliance/generate_fixtures.R || true
+          fi
+
+      - name: Run bitwise compliance (OMRF configs only)
+        run: Rscript tests/compliance/test_compliance.R
+

.gitignore

@@ -3,6 +3,7 @@
 **/.Rhistory
 .RData
 .Ruserdata
+.vscode/
 src/*.o
 src/*.so
 src/*.dll
@@ -18,4 +19,5 @@ src/sources.mk
 docs/*
 /doc/
 /inst/doc/
-dev/plans/
+dev/
+/paper/

DESCRIPTION

@@ -17,9 +17,10 @@ Authors@R: c(
   )
 Maintainer: Maarten Marsman <[email protected]>
 Description: Bayesian estimation and variable selection for Markov random field
-    models of networks of binary, ordinal, and continuous variables. Supports
-    Gaussian graphical models, multi-group comparison via 'bgmCompare', and
-    provides simulation, prediction, and missing data imputation.
+    models of networks of binary, ordinal, continuous, and mixed variables.
+    Supports ordinal MRFs, Gaussian graphical models, mixed MRFs combining
+    discrete and continuous variables, multi-group comparison via 'bgmCompare',
+    and provides simulation, prediction, and missing data imputation.
 Copyright: Includes datasets 'ADHD' and 'Boredom', which are licensed under CC-BY 4. See individual data documentation for license and citation.
 License: GPL (>= 2)
 URL: https://Bayesian-Graphical-Modelling-Lab.github.io/bgms/, https://github.com/Bayesian-Graphical-Modelling-Lab/bgms
@@ -39,6 +40,7 @@ LinkingTo:
     RcppParallel, 
     dqrng, 
     BH
+Roxygen: list(markdown = TRUE)
 RoxygenNote: 7.3.3
 Depends: 
     R (>= 3.5)

NAMESPACE

@@ -4,15 +4,15 @@ S3method(coef,bgmCompare)
 S3method(coef,bgms)
 S3method(extract_arguments,bgmCompare)
 S3method(extract_arguments,bgms)
-S3method(extract_category_thresholds,bgmCompare)
-S3method(extract_category_thresholds,bgms)
 S3method(extract_ess,bgmCompare)
 S3method(extract_ess,bgms)
 S3method(extract_group_params,bgmCompare)
 S3method(extract_indicator_priors,bgmCompare)
 S3method(extract_indicator_priors,bgms)
 S3method(extract_indicators,bgmCompare)
 S3method(extract_indicators,bgms)
+S3method(extract_main_effects,bgmCompare)
+S3method(extract_main_effects,bgms)
 S3method(extract_pairwise_interactions,bgmCompare)
 S3method(extract_pairwise_interactions,bgms)
 S3method(extract_posterior_inclusion_probabilities,bgmCompare)
@@ -40,6 +40,7 @@ export(extract_ess)
 export(extract_group_params)
 export(extract_indicator_priors)
 export(extract_indicators)
+export(extract_main_effects)
 export(extract_pairwise_interactions)
 export(extract_pairwise_thresholds)
 export(extract_posterior_inclusion_probabilities)

NEWS.md

@@ -2,6 +2,7 @@
 
 ## New features
 
+* Mixed MRF models: `bgm()` accepts a per-variable `variable_type` vector that mixes `"ordinal"`, `"blume-capel"`, and `"continuous"` types to estimate networks with both discrete and continuous variables. `simulate.bgms()` and `predict.bgms()` also support mixed models.
 * Gaussian graphical models (GGM): `bgm(x, variable_type = "continuous")` fits a GGM with Bayesian edge selection. Pairwise effects are partial correlations from the precision matrix.
 * Missing data imputation: `na_action = "impute"` integrates over missing values during MCMC sampling for both ordinal and continuous models.
 

R/RcppExports.R

@@ -25,6 +25,10 @@ compute_conditional_probs <- function(observations, predict_vars, pairwise, main
     .Call(`_bgms_compute_conditional_probs`, observations, predict_vars, pairwise, main, num_categories, variable_type, baseline_category)
 }
 
+compute_conditional_mixed <- function(x_observations, y_observations, predict_vars, Kxx, Kxy, Kyy, mux, muy, num_categories, variable_type, baseline_category) {
+    .Call(`_bgms_compute_conditional_mixed`, x_observations, y_observations, predict_vars, Kxx, Kxy, Kyy, mux, muy, num_categories, variable_type, baseline_category)
+}
+
 sample_omrf_gibbs <- function(num_states, num_variables, num_categories, pairwise, main, iter, seed) {
     .Call(`_bgms_sample_omrf_gibbs`, num_states, num_variables, num_categories, pairwise, main, iter, seed)
 }
@@ -45,10 +49,22 @@ run_ggm_simulation_parallel <- function(pairwise_samples, main_samples, draw_ind
     .Call(`_bgms_run_ggm_simulation_parallel`, pairwise_samples, main_samples, draw_indices, num_states, num_variables, means, nThreads, seed, progress_type)
 }
 
+sample_mixed_mrf_gibbs <- function(num_states, Kxx_r, Kxy_r, Kyy_r, mux_r, muy_r, num_categories_r, variable_type_r, baseline_category_r, iter, seed) {
+    .Call(`_bgms_sample_mixed_mrf_gibbs`, num_states, Kxx_r, Kxy_r, Kyy_r, mux_r, muy_r, num_categories_r, variable_type_r, baseline_category_r, iter, seed)
+}
+
+run_mixed_simulation_parallel <- function(mux_samples, kxx_samples, muy_samples, kyy_samples, kxy_samples, draw_indices, num_states, p, q, num_categories, variable_type_r, baseline_category, iter, nThreads, seed, progress_type) {
+    .Call(`_bgms_run_mixed_simulation_parallel`, mux_samples, kxx_samples, muy_samples, kyy_samples, kxy_samples, draw_indices, num_states, p, q, num_categories, variable_type_r, baseline_category, iter, nThreads, seed, progress_type)
+}
+
 sample_ggm <- function(inputFromR, prior_inclusion_prob, initial_edge_indicators, no_iter, no_warmup, no_chains, edge_selection, seed, no_threads, progress_type, edge_prior = "Bernoulli", beta_bernoulli_alpha = 1.0, beta_bernoulli_beta = 1.0, beta_bernoulli_alpha_between = 1.0, beta_bernoulli_beta_between = 1.0, dirichlet_alpha = 1.0, lambda = 1.0, na_impute = FALSE, missing_index_nullable = NULL) {
     .Call(`_bgms_sample_ggm`, inputFromR, prior_inclusion_prob, initial_edge_indicators, no_iter, no_warmup, no_chains, edge_selection, seed, no_threads, progress_type, edge_prior, beta_bernoulli_alpha, beta_bernoulli_beta, beta_bernoulli_alpha_between, beta_bernoulli_beta_between, dirichlet_alpha, lambda, na_impute, missing_index_nullable)
 }
 
+sample_mixed_mrf <- function(inputFromR, prior_inclusion_prob, initial_edge_indicators, no_iter, no_warmup, no_chains, edge_selection, seed, no_threads, progress_type, edge_prior = "Bernoulli", beta_bernoulli_alpha = 1.0, beta_bernoulli_beta = 1.0, beta_bernoulli_alpha_between = 1.0, beta_bernoulli_beta_between = 1.0, dirichlet_alpha = 1.0, lambda = 1.0, sampler_type = "mh", target_acceptance = 0.80, max_tree_depth = 10L, num_leapfrogs = 100L, na_impute = FALSE, missing_index_discrete_nullable = NULL, missing_index_continuous_nullable = NULL) {
+    .Call(`_bgms_sample_mixed_mrf`, inputFromR, prior_inclusion_prob, initial_edge_indicators, no_iter, no_warmup, no_chains, edge_selection, seed, no_threads, progress_type, edge_prior, beta_bernoulli_alpha, beta_bernoulli_beta, beta_bernoulli_alpha_between, beta_bernoulli_beta_between, dirichlet_alpha, lambda, sampler_type, target_acceptance, max_tree_depth, num_leapfrogs, na_impute, missing_index_discrete_nullable, missing_index_continuous_nullable)
+}
+
 sample_omrf <- function(inputFromR, prior_inclusion_prob, initial_edge_indicators, no_iter, no_warmup, no_chains, edge_selection, sampler_type, seed, no_threads, progress_type, edge_prior = "Bernoulli", na_impute = FALSE, missing_index_nullable = NULL, beta_bernoulli_alpha = 1.0, beta_bernoulli_beta = 1.0, beta_bernoulli_alpha_between = 1.0, beta_bernoulli_beta_between = 1.0, dirichlet_alpha = 1.0, lambda = 1.0, target_acceptance = 0.8, max_tree_depth = 10L, num_leapfrogs = 10L, pairwise_scaling_factors_nullable = NULL) {
     .Call(`_bgms_sample_omrf`, inputFromR, prior_inclusion_prob, initial_edge_indicators, no_iter, no_warmup, no_chains, edge_selection, sampler_type, seed, no_threads, progress_type, edge_prior, na_impute, missing_index_nullable, beta_bernoulli_alpha, beta_bernoulli_beta, beta_bernoulli_alpha_between, beta_bernoulli_beta_between, dirichlet_alpha, lambda, target_acceptance, max_tree_depth, num_leapfrogs, pairwise_scaling_factors_nullable)
 }

R/bgm.R

@@ -1,16 +1,18 @@
 #' Bayesian Estimation or Edge Selection for Markov Random Fields
 #'
 #' @description
-#' The \code{bgm} function estimates the pseudoposterior distribution of
-#' category thresholds (main effects) and pairwise interaction parameters of a
-#' Markov Random Field (MRF) model for binary and/or ordinal variables.
-#' Optionally, it performs Bayesian edge selection using spike-and-slab
-#' priors to infer the network structure.
+#' The \code{bgm} function estimates the pseudoposterior distribution of the
+#' parameters of a Markov Random Field (MRF) for binary, ordinal, continuous,
+#' or mixed (discrete and continuous) variables. Depending on the variable
+#' types, the model is an ordinal MRF, a Gaussian graphical model (GGM), or a
+#' mixed MRF. Optionally, it performs Bayesian edge selection using
+#' spike-and-slab priors to infer the network structure.
 #'
 #' @details
-#' This function models the joint distribution of binary and ordinal variables
-#' using a Markov Random Field, with support for edge selection through Bayesian
-#' variable selection. The statistical foundation of the model is described in
+#' This function models the joint distribution of binary, ordinal, continuous,
+#' or mixed variables using a Markov Random Field, with support for edge
+#' selection through Bayesian variable selection. The statistical foundation
+#' of the model is described in
 #' \insertCite{MarsmanVandenBerghHaslbeck_2025;textual}{bgms}, where the ordinal
 #' MRF model and its Bayesian estimation procedure were first introduced. While
 #' the implementation in \pkg{bgms} has since been extended and updated (e.g.,
@@ -172,8 +174,11 @@
 #'   Blume–Capel variables, all categories are retained.
 #'
 #' @param variable_type Character or character vector. Specifies the type of
-#'   each variable in \code{x}. Allowed values: \code{"ordinal"} or
-#'   \code{"blume-capel"}. Binary variables are automatically treated as
+#'   each variable in \code{x}. Allowed values: \code{"ordinal"},
+#'   \code{"blume-capel"}, or \code{"continuous"}. A single string applies
+#'   to all variables. A per-variable vector that mixes discrete
+#'   (\code{"ordinal"} / \code{"blume-capel"}) and \code{"continuous"}
+#'   types fits a mixed MRF. Binary variables are automatically treated as
 #'   \code{"ordinal"}. Default: \code{"ordinal"}.
 #'
 #' @param baseline_category Integer or vector. Baseline category used in
@@ -212,6 +217,21 @@
 #'   score endpoints \eqn{(-b, m-b)}.
 #'   Default: \code{FALSE}.
 #'
+#' @param pseudolikelihood Character. Specifies the pseudo-likelihood
+#'   approximation used for mixed MRF models (ignored for pure ordinal or
+#'   pure continuous data). Options:
+#'   \describe{
+#'     \item{\code{"conditional"}}{Conditions on the observed continuous
+#'       variables when computing the discrete full conditionals. Faster
+#'       because the discrete pseudo-likelihood does not depend on the
+#'       continuous precision matrix.}
+#'     \item{\code{"marginal"}}{Integrates out the continuous variables,
+#'       giving discrete full conditionals that account for induced
+#'       interactions through the continuous block. More expensive per
+#'       iteration.}
+#'   }
+#'   Default: \code{"conditional"}.
+#'
 #' @param main_alpha,main_beta Double. Shape parameters of the
 #'   beta-prime prior for threshold parameters. Must be positive. If equal,
 #'   the prior is symmetric. Defaults: \code{main_alpha = 0.5} and
@@ -307,16 +327,26 @@
 #' Main components include:
 #' \itemize{
 #'   \item \code{posterior_summary_main}: Data frame with posterior summaries
-#'     (mean, sd, MCSE, ESS, Rhat) for category threshold parameters.
+#'     (mean, sd, MCSE, ESS, Rhat) for main-effect parameters.
+#'     For OMRF models these are category thresholds;
+#'     for mixed MRF models these are discrete thresholds and
+#'     continuous means. \code{NULL} for GGM models (no main effects).
+#'   \item \code{posterior_summary_quadratic}: Data frame with posterior
+#'     summaries for the precision matrix diagonal. Present for GGM and
+#'     mixed MRF models; \code{NULL} for OMRF models.
 #'   \item \code{posterior_summary_pairwise}: Data frame with posterior
 #'     summaries for pairwise interaction parameters.
 #'   \item \code{posterior_summary_indicator}: Data frame with posterior
 #'     summaries for edge inclusion indicators (if \code{edge_selection = TRUE}).
 #'
-#'   \item \code{posterior_mean_main}: Matrix of posterior mean thresholds
-#'     (rows = variables, cols = categories or parameters).
+#'   \item \code{posterior_mean_main}: Posterior mean of main-effect
+#'     parameters. \code{NULL} for GGM models. For OMRF: a matrix
+#'     (p x max_categories) of category thresholds. For mixed MRF: a list
+#'     with \code{$discrete} (threshold matrix) and \code{$continuous}
+#'     (q x 1 matrix of means).
 #'   \item \code{posterior_mean_pairwise}: Symmetric matrix of posterior mean
-#'     pairwise interaction strengths.
+#'     pairwise interaction strengths. For GGM and mixed MRF models the
+#'     precision matrix diagonal is included on the matrix diagonal.
 #'   \item \code{posterior_mean_indicator}: Symmetric matrix of posterior mean
 #'     inclusion probabilities (if edge selection was enabled).
 #'
@@ -409,6 +439,7 @@ bgm = function(
   display_progress = c("per-chain", "total", "none"),
   seed = NULL,
   standardize = FALSE,
+  pseudolikelihood = c("conditional", "marginal"),
   verbose = getOption("bgms.verbose", TRUE),
   interaction_scale,
   burnin,
@@ -481,7 +512,8 @@ bgm = function(
     cores = cores,
     seed = seed,
     display_progress = display_progress,
-    verbose = verbose
+    verbose = verbose,
+    pseudolikelihood = pseudolikelihood
   )
 
   raw = run_sampler(spec)