Tools for Bayesian workflows
Although individual workdflows may vary depending on the requirements of the given project, there exists a range of standard methods that are recommended to be part of cognitive modelling workflows to ensure that analysis is completed correctly. These are introduced in more detail by various sources: see (Hess et al., 2025; Wilson & Collins, 2019; Lee & Wagenmakers, 2014) for examples. This includes prior and posterior checks, which ensure that the model produces appropriate behavioural, precision analysis and parameter recovery, which ensure that parameters can be accurately estimated from the data, and model comparison, which allows to compare models in terms of how well they describe the data. It also includes basic diagnostic checks, to ensure that the MCMC sampling has been succesful. ActionModels provides ready funtions to implement these methods, which are described in the following sections.
Chain diagnostics
Chain diagnostics with Turing
#TODO: diagnostics with Turing's own output
This includes the rhat value, which indicates whether the chains have converged, and which should be close to 1 for all parameters. It also includes the ess_bulk
and ess_tail
values, which indicate the effective sample size of the chains.
Chain diagnostics with ArviZ
#TODO: use ArviZ
Looking for parameter correlations
#TODO: Make plot (or use ArviZ)
Model comparison
#TODO: use ArviZ
Predictive checks
#TODO: finish functions
Prior predictive checks
Posterior predictive checks
Precision analysis
#TODO: finish functions
Parameter recovery
Model recovery
Experiment tuning
This page was generated using Literate.jl.