Internal Refactoring: of a) pyfixest FormulaParser and b) full support of narwhals #1182
Description
Background
To support fixest formula innovations, PyFixest includes quite a lot of ugly regex code internally.
We would like to substantially clean up the logic by moving much more work to formulaic.
@leostimpfle has provided an initial PR here.
Additionally, we would like to offer full support for narwhals, which is now possible as formulaic supports it end-to-end!
This means that instead of converting from polars to pandas internally, we would like to fully compute on polars data frames in case users provide polars, and also return polars dfs and series.
To do
- Consider moving to
formulaicIV syntax, as this should significantly decrease maintenance burden for us. For details, see this issue Implementformulaic's multistage syntax #1125 - Implement all features of fixest i() syntax by relying on formulaic: Mimic
fixest::i()by relying on formulaic stateful transforms #782 - Overhaul the documentation for multiple estimation syntax and all tricks. In particular, provide benchmarks on the performance gains! Document options for multiple dependent variables #1116
- Implement multiverse stepwise estimation syntax
mvsw#1136 - This should fix some open issues: Bug:
predict()method fails with new data and pyfixest special syntax -i(),C(),var1^var2operators #684 andi()does not accept two categorical terms #921 - IV with multiple endogenous variables #1117
- Support syntax for varying slopes.
- Fully support
narwhals.
Related issues
- Support Native Narwhals #967
- narwhals = Pandas + polars + ... #533
- Refactor formula parsing #1118
- Implement
formulaic's multistage syntax #1125 - Document options for multiple dependent variables #1116
- Bug:
predict()method fails with new data and pyfixest special syntax -i(),C(),var1^var2operators #684 i()does not accept two categorical terms #921- Sum of variables not parsed correctly in multiple estimation syntax #1137
- Advanced binning functionality for
fixest::isyntax #1130