can Stabl support multi-class logistic?

[dorothyzh]

If I have three class A B C, how could I use mullti-class or multi nomial logsti or other model in the Stabl?

[dorothyzh]

It seems like the Stabl does not support multi class....

[xavdurand]

Hello @dorothyzh ,

Stabl supports Multi-class if the base_estimator model supports multiclass.
The default base_estimator is for binary purpose only, you have to change the default one.
I hope it helps you.

[dorothyzh]

how to change the default one? thanks, is there any demo notebook?

[xavdurand]

hello @dorothyzh ,

First, define your base_estimator:

estimator = LogisticRegression(multi_class='ovr', penalty="l1", class_weight="balanced", solver="liblinear")

The logistic regression support multiclass thanks to the parameter multi_class="ovr".

Then, you set the base estimator of the Stabl:

stabl = Stabl(
    base_estimator=estimator,
    n_bootstraps=500,
    artificial_type=artificial_type,
    artificial_proportion=1.,
    replace=False,
    lambda_grid={"C": np.linspace(0.01, 1, 30)},
)

Note: there is a warning if the sklearn version is ~1.5.