search

NASA-Openscapes / py-rocket

Rocker (R Docker stack) via Jupyter lab
MIT License
2 stars 1 forks source link

Add reticulate and earthdatalogin R packages to base image #4

Open ateucher opened 5 months ago

ateucher commented 5 months ago

Related to #5

eeholmes commented 5 months ago

@ateucher See packages here that Carl added which includes these plus a few others. This is my update to py-rocket.

https://github.com/nmfs-opensci/container-images/tree/main/images/py-rocket-geospatial

ateucher commented 5 months ago

Brilliant, thanks @eeholmes - those are pretty much the packages I was thinking about! Can you tell me again why you use a base image with rocker/verse + the conda stuff, then you add the geospatial via /rocker_scripts/install_geospatial.sh, rather than using the rocker/geospiatal image as the base?

ateucher commented 5 months ago

I do probably think it's a good idea to replicate the conda python environment from corn, as I am sure there is likely a set of users who will jump between both. Also from an administrative standpoint, managing one kind of python environment is easier. I am sure there must be a way to pull the conda setup from corn more efficiently than manual copy/paste.

eeholmes commented 5 months ago

The conda version is bigger, which doesn't matter for JupyterHub so much but does for Codespaces, Binder etc.

It is possible that using the conda-lock file from the Openscapes/python image would allow us to completely duplicate the Python environment wo having such a big image.

I am not sure of the implications of using venv w a conda-lock file on using the py-rocket as a base image for other Py-R images.

eeholmes commented 5 months ago

Also this is relevant: Pangeo design goals https://pangeo-docker-images.readthedocs.io/en/latest/topic/design.html