Stop wrangling and start analyzing your MCMC samples!
This R package was born out of my own frustrations with dealing with MCMC samples. I was wasting too much time wrangling the MCMC samples in order to get what I needed for my analyses.
Sometimes the MCMC samples are needed in their flat form, e.g. in order to calculate convergence statistics. Sometimes the samples are needed in their original dimensions, e.g. in order to calculate the predicted values or other functions of the parameters. Sometimes, the samples need to be seperated by chain, e.g. convergence diagnostics. Sometimes the samples needed to be pooled across chains, e.g. calculating quantiles.
This package aims to do one thing: make storing, accessing, and manipulating MCMC samples easier and faster. It does not include convergence statistics, or plotting functions; although it will make applying such functions to samples easier.
It defines generic functions to access the MCMC samples in a common manner, regardless of how they are stored. This makes is possible to store samples in a variety of formats, while accessing them with the same functions. Currently, this package only contains a class for storing samples in memory. However, SQLite and other backends are planned.
Use devtools to install mcmcdb from github.
library(devtools) install_github(c(r-"checker", "mcmcdb"), "jrnold")
McmcdbWide objects can be created directly, however it is more
likely that MCMC samples are either written to disk or in an R object
of another format. The function
McmcdbWide will create new
McmcdbWide from matrix,
mcmc.list objects. The
mcmcdb_wide_from_stan will create a
from the csv files output by a Stan command line program.