polars -> narwhals

[s3alfisc]

Full migration from all code parts that rely on polars to narwhals. Drops polars dependency.

[s3alfisc]

Hi @MarcoGorelli ,I am trying to fully deprecate the polars dependency in favor of narwhals. I am using pl.count() for which narwhals does not seem to support a direct equivalent. Is that correct? If yes, is there any other way to rewrite

    agg_expressions.append(pl.count(depvars[0]).alias("count"))

to narwhals supported syntax?

[s3alfisc]

Actually, this isn't even true, of course there is a count in narwhals: link.

[s3alfisc]

This should work now =)

[codecov]

Codecov Report

Attention: Patch coverage is 71.42857% with 12 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
pyfixest/estimation/feols_compressed_.py	72.97%	10 Missing ⚠️
pyfixest/estimation/feols_.py	60.00%	2 Missing ⚠️

Flag	Coverage Δ
core-tests	77.32% <71.42%> (-0.59%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
pyfixest/estimation/feols_.py	83.20% <60.00%> (-0.08%)	⬇️
pyfixest/estimation/feols_compressed_.py	69.23% <72.97%> (-14.21%)	⬇️

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests

[s3alfisc]

Unit tests in tests/test_feols_compressed.py are now very slow - looks like I lose a lot of performance, likely by doing something that is not recommended 🤔

[s3alfisc]

I think I completely misunderstood how narwhals works - after all, I now realized that narwhals does not copy code from pandas to polars and runs polars under the hood (not sure where I got this from, maybe because all the R translators of dplyr do this?), but simply translates code! So the performance drop came from running all computations via pandas instead of polars 🤦