Best way to handle managing multiple environments in the same project

[Mathemmagician]

Question

In a research data science team setting, different members might prefer to use their personal environments when doing research, but also at times be able to quickly switch between others'. What is the best way of structuring this in general? I.e. should we create an envs folder with subdirectories for each env? Assuming there is single one 'data-science-repo'

Platform

No response

Version

No response

[ivanbelenky]

I would say that inconsistencies don't look good, but if deemed completely necessary, UV_PROJECT_ENVIRONMENT is your friend. You can probably create a small bash script to switch between venvs in tandem with preserving state using group dependencies in your pyproject.toml.

I would also suggest/emphasize that different dependencies should be based on functionality and not user preference.

[konstin]

One option is using uv run --with <...> to test the project plus separate dependencies. Another option is manual venv management with uv venv <name> && . name/bin/activate && uv pip install <...>

[Mathemmagician]

I should have also mentioned using conda works great in this case, as environments aren't tied to a project (but the repo does have a envs folder with env .yml files). Maybe I should have framed the question as - is there a conda way of using uv?

And again, this is not a production level repository, rather an exploratory research one

[notatallshaw]

is there a conda way of using uv?

I think you're looking for the highly requested #1495

I'm pretty sure the answer at the moment is not really. At my work I still use conda to set up multiple global environments, and then uv to install Python packages in those conda environments.

[Mathemmagician]

I think that's it! Thank you