ValueError on Multiple Calls to fit_conformalize in CrossConformalRegressor

[CyrilJl]

This is not a bug, but a question regarding the following code:

from sklearn.datasets import make_regression
from mapie.regression import CrossConformalRegressor
from sklearn.linear_model import Ridge

model = CrossConformalRegressor(estimator=Ridge())

X, y = make_regression()
model.fit_conformalize(X, y)

another_X, another_y = make_regression()
model.fit_conformalize(another_X, another_y)

When executed, this code raises a ValueError: fit_conformalize method already called. MAPIE does not currently support calling fit_conformalize several times.

I couldn't find any information about this behavior in the documentation. It surprised me as it differs from the behavior in versions prior to v1 and also from scikit-learn, which allows the object to be reused. I'm curious about the reasoning behind this implementation. Additionally, I'm wondering if, for my needs, I have no choice but to create a new instance of the Mapie regressor each time I use it on a new dataset.

[Valentin-Laurent]

Hello @CyrilJl. It is indeed something we're considering implementing, so thank you for the feedback. We realised there are unwanted side effects in some cases, that's why we decided not to support re-fitting/re-fit_conformalizing for now.

Regarding your specific needs, you're right, you can instanciate a new Mapie regressor.

[Valentin-Laurent]

@CyrilJl I'm reopening the issue, it will be interesting to see if more users have this need so we can prioritise it accordingly, and to give visibility to potential contributors.