Movatterモバイル変換

batchmix

Downloads from CRAN in the past month:

Semi-supervised and unsupervised Bayesian mixture models thatsimultaneously infer the cluster/class structure and a batch correction.Densities available are the multivariate normal and the multivariate t.The model sampler is implemented in C++. This package is aimed atanalysis of low-dimensional data generated across several batches. SeeColeman etal. (2022) for details of the model.

Advice on using the package

The main functions a user should be aware of arerunMCMCChains,plotLikelihoods,plotAcceptanceRates,continueChains andprocessChains.

Parameters are sampled using Metropolis-Hastings so checking that theacceptance rate is important. We recommend aiming for acceptance ratesbetween 0.1 and 0.5 for the class and batch means and batch scales(\(\mu_k\),\(m_b\) and\(S_b\) respectively). In our testing, anacceptance rate of at least 0.4 for the class covariance matrices tendedto suggest the sampler is exploring well, but smaller values werefrequently associated with poor behaviour. The degrees of freedom tendto have very high acceptance rates in our testing regardless of thesampling window.

We recommend running a small number of chains for a small number ofiterations to assess the acceptance rates before committing thecomputational resourcces to run a full analysis.

For an example of a workflow please see the short vignette.