Global Options

[s3alfisc]

Allows to set global options for estimation. Would close #766.

Example:

%load_ext autoreload
%autoreload 2

import pyfixest as pf 
data = pf.get_data()

pf.set_option(
    data = pf.get_data().dropna(), 
    vcov = {"CRV1": "f1"}, 
    weights = "weights", 
    demeaner_backend = "rust"
)

# if arg not explicity provided, looks for and uses global default
fit1 = pf.feols(fml = "Y ~ X1 + X2 | f2", vcov = "hetero")
fit2 = pf.feols(fml = "Y ~ X1 | f1 + f2")
fit3 = pf.feols(fml = "Y ~ X1 + X2 ")
pf.etable([fit1, fit2, fit3],)

Is this a feature we'd like to have? Are there any risks? @apoorvalal @juanitorduz

[apoorvalal]

I can see the appeal of something like this for interactive use, but also it introduces a lot of potentially consequential hidden state that makes reproducibility a pain and should generally be discouraged.
What if we did the intermediate thing where we let one set a bunch of preferences in a dict and pass the entire dict to each model fit call and unpack it inside feols with **kwargs ? That's roughly how scikit learners with many hyperparameters work right?

[juanitorduz]

I agree with @apoorvalal ! Having these hidden variables can be very painful to debug. the intermediate suggestion seems reasonable :)