antoine-sachet / rocker-ML

Dockerfiles for ML with R - built on top of Rocker

R docker images for statistical modelling

The rocker project provides a fantastic stack of docker images built from debian. This repository extends the rocker images to provide machine learning / statistical modelling capabilities.

FROM rocker/verse

All these docker images are based on rocker/verse which already contains:

• r-base
• rstudio
• tidyverse
• shiny
• publishing tools (pandoc, tex)

Version-specific branches are based on the relevant rocker/verse version (e.g. asachet/rocker-ml:3.6.1 is based on rocker/verse:3.6.1 which uses R 3.6.1).

rocker-ml

The rocker-ml image adds to rocker/verse:

• ggplot2 utilities such as scales, ggridges, GGally, RColorBrewer, Viridis
• Misc tools such as foreach, Hmisc, doParallel, openxlsx
• Modelling tools:
  • caret and related (ensembleR, MLmetrics)
  • statistical utilities like MASS, car, fitdistrplus
  • A large number of modelling packages including gam, lme4, e1071, kernlab, survival, elasticnet, randomForest, ranger, ...
  • Open the dockerfile to see the complete list.

rocker-stan

The rocker-stan image adds Stan to rocker-ml:

• rstan, rstantools, rstanarm, shinystan, bayesplot, brms, tidybayes

rocker-tidymodels

The rocker-tidymodels image adds tidymodels to rocker-stan. Note that rstanarm is a requirement for tidyposterior hence the stack.

rocker-keras

The rocker-keras image adds to rocker/tidymodels:

• TensorFlow (latest CPU version) and Keras (latest version)
• R packages tensorflow, keras and tfestimators

See https://tensorflow.rstudio.com/keras/ for more info.