ENH: Add datalad installation via conda

[yarikoptic]

It will also install git-annex which datalad uses for management of large files.

Since it is just a "tool", I have not used --freeze-installed to benefit from possible upgrades etc.

It could also be installed from stock ubuntu, but would be an older version. Also backports are available from NeuroDebian, but I have decided to just go with conda installation for now.

Closes #50

What kind of change does this PR introduce?

• [ ] Bugfix
• [x] New Feature
• [ ] Feature Improvment
• [ ] Refactoring
• [ ] Documentation
• [ ] Other, please describe:

Description:

Checklist:

• [x] I have read the CONTRIBUTING document.
• [ ] My changes don't require a change to the documentation, or if they do, I've added all required information.

[yarikoptic]

Oh, I guess I should add to documentation. At large it should be just a reference to http://handbook.datalad.org/ and may be http://handbook.datalad.org/en/latest/code_from_chapters/usecase_ml_code.html and http://handbook.datalad.org/en/latest/beyond_basics/101-168-dvc.html in particular

[lukasmasuch]

@yarikoptic Thanks for the pull request! I took a closer look at datalab for the 0.12 release. Unfortunately, the installation is too big for a single tool to be preinstalled inside the main workspace flavor:

• conda install (409MB):

  

• apt install (390MB):

  

A few other options:

1. Build and publish your own datalad flavor:

   
   # Extend from any of the workspace versions/flavors
   FROM mltooling/ml-workspace:0.12.1

Run you customizations, e.g.

RUN \

Install datalad

apt-get update && \
apt-get install -y datalad && \
# Cleanup Layer - removes unneccessary cache files
clean-layer.sh

2. Use an install method with a smaller footprint. However, since most space is used by git-annex, this might not be possible.
3. Add a tool installer script (e.g. see [these scripts](https://github.com/ml-tooling/ml-workspace/tree/main/resources/tools)). However, datalad is quite easy to install, so it might not need an additional installer script. 

[yarikoptic]

conda install (409MB) -- is that in addition to what you already have installed otherwise anyways ( indeed would be too big then :-/) ? majority should be common python packages (besides git-annex itself which is written in haskell but it should not pull too much). I can look into it later to see where the hit comes from and see what could be done.

[lukasmasuch]

Yes, it's in addition. But there are some pinned dependencies that are forced to downgrade/upgrade... which might blow up the total required space as well. To see what's actually installed/removed/downgraded, you can just build this Dockerfile:

# Extend from any of the workspace versions/flavors
FROM mltooling/ml-workspace:0.12.1

# Run you customizations, e.g.
RUN \
    # Install datalad
    apt-get update && \
    apt-get install -y datalad && \
    # Cleanup Layer - removes unneccessary cache files
    clean-layer.sh

[github-actions[bot]]

This PR is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 14 days