[Medium] Aligning better again with https://github.com/Quantco/pypackage-cookiecutter

[quantco]

[Medium] Aligning better again with https://github.com/Quantco/pypackage-cookiecutter

[ElizabethSantorellaQC]

Sounds good to me. I think we just use Docker because we're more familiar with it.

[tbenthompson]

Earlier, when we were using a bunch of other packages that had funky dependencies, Docker was more useful: tensorflow and h2o were the two main ones that required specific OS-level dependencies that I don't think conda handles. H2O needed Java.

I've just been bitten by dependency/installation issues working in a bigger group so many times that I default to using docker because it solves almost all those issues.

When we're dealing just with the sklearn package, I don't care between conda and docker.

[xhochy]

Earlier, when we were using a bunch of other packages that had funky dependencies, Docker was more useful: tensorflow and h2o were the two main ones that required specific OS-level dependencies that I don't think conda handles. H2O needed Java.

The main point of conda is to handle OS-level dependencies. Java is available as the openjdk package. If other things are missing, feel free to reach out to me, these things are quickly fixed in most cases.

When we're dealing just with the sklearn package, I don't care between conda and docker.

Still, if you would care, I'm happy to add a Dockerfile to the setup. It shouldn't just cater for my needs but for the general QC audience.

[tbenthompson]

The main point of conda is to handle OS-level dependencies. Java is available as the openjdk package. If other things are missing, feel free to reach out to me, these things are quickly fixed in most cases.

Learning things all the time! I didn't know how mature the conda-forge stack is! Awesome.

Still, if you would care, I'm happy to add a Dockerfile to the setup. It shouldn't just cater for my needs but for the general QC audience.

I don't think that's necessary.