The iowa package implements modular reinforcement learning models of the Iowa gambling task. Model components are implemented in Stan, and are compiled into cmdstan models which are then made available internally to other packages.
The package implements the simulation and fitting of models constructed by mixing and matching various utility, updating, and temperature functions; and is designed to be relatively extensible by allowing users to implement custom model components.
In addition to simulating the performance of custom models, iowa also allows model fitting either by maximum likelihood / maximum a posteriori estimation, or by full posterior sampling. Currently, only single subject fitting is supported, but support for full hierarchical Bayesian fitting is a strong priority.
Installation
Models in iowa are pre-compiled using cmdstanr, which must be installed alongside cmdstan. Both can be installed within R using
remotes::install_github("stan-dev/cmdstanr")
cmdstanr::install_cmdstan()iowa can then be installed directly from its repository:
devtools::install_github('areshenk-rpackages/iowa', type = 'source')