Add `minimal` environment option

[s3alfisc]

Should we add a "minimal" environment that allows to use core functionality but drops many dependencies only required for select operations?

Currently, the "main" env has the following dependencies:

• "scipy>=1.6",
• "formulaic>=1.1.0",
• "pandas>=1.1.0",
• "numba>=0.58.0",
• "seaborn>=0.13.2",
• "tabulate>=0.9.0",
• "tqdm>=4.0.0",
• "great-tables>=0.10.0",
• "numpy>=1.19.0",
• "narwhals>=1.13.3",
• "joblib>=1.4.2,<2",

Instead, the "minimal" environment would contain only

• "scipy>=1.6",
• "formulaic>=1.1.0",
• "pandas>=1.1.0",
• "numba>=0.58.0",
• "numpy>=1.19.0",
• "narwhals>=1.13.3",

As a result, plotting methods would not work (no seaborn), etable and summary would not work (no tabulate, great-tables), some inference methods that depend on tqdm and joblib might not work (ritest, ccv, gelbach).

Is there a use case for such an environment?

[elchorro]

Not sure how complicated this would be, but I'd say that it would be useful. I mentioned elsewhere that I want to bring in pyfixest to my company's internal codebase: no conda/pip, all dependencies have to be carefully managed.

I'd definitely start with the minimal distribution, if it was available.

[s3alfisc]

Thanks for the feedback @elchorro! I think this should be fairly easy to do, we'd simply have to add the "small" env to the pyproject - the summarization and report functions live in separate standalone scripts, and so does most of the core implementation of outside methods. Though I'd have to verify this. Beyond a bit of dependency management in the script, I think no code adjustments would be needed as whenever a dependency is missing, pyfixest should fail quite loudly?

[s3alfisc]

Ok just tested this a bit and it will be more work than I had hoped for (in the sense that it's not a 60 minutes effort) ... it will require to reorganize import statements by quite a lot. I have to tinker about this a bit & have to come up with a plan - but I definitely would love to help you to get pyfixest into prod @elchorro =)

[s3alfisc]

Actually, not that much work after all, apparently =) Very MVP PR here: #830

[apoorvalal]

When adding pyfixest to a uv project, it attempts to install statsmodels and lets-plot, which is puzzling. Neither are in the 'core' set of dependencies in pyproject.toml. Any idea what's going on?

[s3alfisc]

This is indeed strange, I'll have to take a closer look. I agree that I'd expect that only what is contained in the default env should be installed by default... only ad hoc idea I have - maybe you're installing an old version? Though I think sm never was a core dependency.