The ggdmc R package provides tools and tutorials for cognitive modelling, supporting both Bayesian and non-Bayesian approaches. Evolving from Dynamic Models of Choice (DMC; Heathcote, Lin, et al., 2018), ggdmc targets challenging hierarchical and likelihood-free problems while remaining friendly for conventional modelling workflows.

Latest release: v0.2.9.0 — expanded hierarchical tools, improved samplers, and better monitoring for model fitting.

---

🚀 What’s new in v0.2.9.0

• Parallel chain instances for hierarchical models
  By default, hierarchical fits now launch three independent chain instances, each running a swarm of chains (≈ 3× the number of parameters) to improve posterior exploration and guard against cross-chain dependence.

• Sampler & monitoring improvements
  More robust defaults and clearer diagnostics for convergence and mixing.

• Richer hierarchical controls
  Cleaner parameter control plus more flexible variability settings for DDM, LBA, and CDM.

• Migration & blocking options
  Enable migration and blocking at subject and/or population levels to speed convergence on complex factorial designs.

---

✨ Key Features

1. Population-based MCMC (pMCMC)
   Multiple interacting chains for efficient exploration of complex posteriors.

2. Multiple samplers, parallelized
   Compare/choose samplers; parallel chain instances provide independence checks and efficiency.

3. Flexible sampling strategies
   Migration operators and blocking at different hierarchy levels.

4. Broader model coverage
   Enhanced support for hierarchical DDM, LBA, and CDM, with clearer variability controls.

---

📦 Installation

CRAN (stable) or source tarball

install.packages("ggdmc")
# Or a specific release tarball:
install.packages("ggdmc_0.2.9.0.tar.gz", repos = NULL, type = "source")

GitHub (development)

# Requires devtools
install.packages("devtools")
devtools::install_github("yxlin/ggdmc")

• Windows: Install Rtools to compile C++.
• macOS: Install Xcode Command Line Tools.
• Linux: Ensure a recent compiler toolchain is available.

🧭 Quick start

library(ggdmc)

# 1) Specify a model (e.g., LBA/DDM/CDM) and priors
# 2) Build a data-model instance (DMI)
# 3) Configure pMCMC (chains, instances, migration/blocking)
# 4) Run sampler, monitor convergence, and examine posterior summaries

🔬 Advanced Example: CDM simulation & hierarchical fitting (click to expand)

```r # Example workflow: build, simulate, fit pkg <- c("ggdmc", "ggdmcPrior", "ggdmcModel", "cdModel") sapply(pkg, require, character.only = TRUE) # Build a CDM model model <- BuildModel( p_map = list(guess1="1", guess2="1", guess3="1", slip1="1", slip2="1", slip3="1"), type = "cdm" ) # Define population priors pop_mean <- c(guess1=.1, guess2=.2, guess3=.3, slip1=.01, slip2=.02, slip3=.03) pop_scale <- c(guess1=.01, guess2=.01, guess3=.01, slip1=.05, slip2=.01, slip3=.01) pop_dist <- ggdmcPrior::BuildPrior( p0=pop_mean, p1=pop_scale, lower=rep(0, model@npar), upper=rep(NA, model@npar), dists=rep("tnorm", model@npar), log_p=rep(FALSE, model@npar) ) # Set CDM models sub_model <- setCDM(model) pop_model <- setCDM(model, population_distribution=pop_dist) # Simulate data dat <- simulate(sub_model, nsim=1000, parameter_vector=pop_mean, nschool=1, seed=123) hdat <- simulate(pop_model, nsim=1000, nschool=32, seed=123) # Build DMIs sub_dmis <- BuildDMI(dat$responses, model, q_matrix=sub_model@q_matrix, rule="DINA") pop_dmis <- BuildDMI(hdat$responses, model, q_matrix=pop_model@q_matrix, rule="DINA") # Priors and initial samples p0 <- rep(0, model@npar); names(p0) <- model@pnames p_prior <- ggdmcPrior::BuildPrior(p0=p0, p1=rep(1.1, model@npar), dist=rep("unif", model@npar), log_p=rep(TRUE, model@npar)) sub_priors <- set_priors(p_prior=p_prior) sub_theta_input <- ggdmc::setThetaInput(nmc=500, pnames=model@pnames) sub_samples <- initialise_theta(sub_theta_input, sub_priors, sub_dmis[[1]]) # Run sampling fits <- StartSampling_subject(sub_dmis[[1]], sub_priors, sub_migration_prob=0.02, thin=2, seed=9032) fit <- RebuildPosterior(fits) hat <- gelman(fit) cat("mpsrf = ", hat$mpsrf, "\n") ```

❓ FAQ

Q1. Installation issues on older R / Microsoft R

• RcppArmadillo may be outdated — install the latest from CRAN.
• If no binary is available, install from source (Rtools on Windows; Xcode on macOS).

Q2. All install methods fail?
Try in this order:

1. CRAN (stable)
2. GitHub (development)
3. Source tarball with toolchain (Rtools/Xcode/GCC/Clang)

---

🔗 Useful Links

• GitHub Source & Issues
• DMC Reference: Heathcote, Lin, et al. (2018)
• Related packages/tutorials: see repository README and vignettes

---

ggdmc continues to evolve to meet research needs in psychology, cognitive science, and education — especially for hierarchical Bayesian and large-scale applications.