pythonr image

[Robinlovelace]

The idea here is to have an R installation on top of a Python-focussed image.

We already have the rocker-rpy image which is the other way around.

Having just done a test, it seems the image can be pretty lean and therefore fast to download, from a recent test:

docker images pythonr-test
# REPOSITORY     TAG       IMAGE ID       CREATED       SIZE
# pythonr-test   latest    e8d70129a508   2 hours ago   1.7GB
docker images rocker-rpy
# REPOSITORY   TAG       IMAGE ID       CREATED         SIZE
# rocker-rpy   latest    a8cf762ffc70   11 months ago   6.27GB

[Robinlovelace]

Reproducible example with the Dockerfile contain the new pythonr image: in incoming Pull Request.

[cboettig]

@Robinlovelace love this, this is what I've been doing as well, building on the Jupyter stack. e.g. in https://github.com/rocker-org/ml. These are more 'batteries included' images (or 'dev containers') for daily work than minimalist images though (especially once you add the CUDA stuff any image gets pretty large; though I'm often in settings where I can benefit from pre-pull cache).

The official Jupyter images are naturally a pretty robust base for python, though are all conda-based. Been meaning to take a go at a uv (or maybe pixi, which uses conda and uv under the hood) based build instead. Anyway, would be keen to compare notes.

You may also already be familiar with the excellent work @eeholmes has done with py-rocket, e.g. https://github.com/nmfs-opensci/py-rocket-base