Cognitive Models logo Cognitive Models

Cognitive Models

ggdmc

This site provides tutorials for the ggdmc R package.

ggdmc is an open-source toolkit for conducting cognitive modelling, supporting both Bayesian and non-Bayesian approaches. Evolving from the Dynamic Models of Choice (DMC; Heathcote, Lin, et al., 2018), the package is designed to address challenging hierarchical and likelihood-free modelling problems, while still accommodating more conventional modelling workflows.

The latest release (v0.2.8.9) introduces expanded functionality, improved sampler options, and enhanced monitoring tools for model fitting.

Key Features

  1. Population-based MCMC (pMCMC) sampling
    ggdmc implements population-based MCMC samplers, which run multiple interacting chains in parallel to improve sampling efficiency.
    This approach provides an alternative to single-chain samplers and can offer better exploration of complex posterior landscapes in some modelling scenarios.

  2. Multiple pMCMC samplers with parallel chain instances
    • ggdmc now offers a broader set of samplers, giving users the flexibility to choose or compare methods for different models.
    • Version 0.2.8.9 introduces a new parallel chain instance concept. By default, hierarchical model fitting launches three independent chain instances. Within each instance, a swarm of chains (three times the number of parameters) is used to enable pMCMC to work effectively.
    • This design addresses the issue of non-independence in traditional pMCMC while also improving computational efficiency.
  3. Flexible migration operator and blocking mechanism
    • Users can enable migration sampling or apply a blocking mechanism at the subject level, population level, or both.
    • This flexibility allows the sampling strategy to be adapted for different model types and factorial designs, improving both convergence and efficiency.
  4. Expanded model support and hierarchical tools (v0.2.8.9)
    • Improved hierarchical model handling with clearer parameter control.
    • More flexible parameter variability settings for DDM and LBA.

Getting Started

Installation

From CRAN or source tarball

install.packages("ggdmc")
install.packages("ggdmc_0.2.8.9.tar.gz", repos = NULL, type = "source")

From GitHub (development version)

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

Windows users: Install Rtools to compile C++ code.
macOS users: Ensure Xcode Command Line Tools are installed.

❓ FAQ

1. Installation fails on Microsoft R

Microsoft R (v3.5.3) may cause issues due to outdated dependencies:

2. All installation methods fail

Try installing from source in the following order: