https://cloud.docker.com/repository/docker/grrrck/rstudio-keras

A Docker Image with RStudio and keras, built on rocker/verse.

This image is not currently built for use with GPUs, but this may be included in the future.

docker run -d -p 8787:8787 -e PASSWORD="SET A PASSWORD" grrrck/rstudio-keras

Open up a localhost:8787 in a browser and login as user rstudio with the password you set in the above command. The rocker/verse image requires that you manually set a password for RStudio; if not set the image will exit quickly and silently and you'll get an error in your browser when connecting to localhost:8787.

If you use the default user (rstudio), keras and tensorflow will be pre-installed in a Python virtual environment in ~/.virtualenvs/r-tensorflow. If you change the default user, you will need to run the following inside the RStudio interface on the image's first start.

library(keras)
install_keras()

If you are running the image on a machine with available GPUs, you can install the GPU-enabled version of TensorFlow by running

library(keras)
install_keras(tensorflow = "gpu")

For more information, TensorFlow for R from RStudio is a great resource.