[BUG] mambular.models.utils.sklearn_parent.SklearnBase.fit() got multiple values for keyword argument 'regression'

[DavidSolan0]

Describe the bug

Calling optimize_hparams() on MambularClassifier (and similarly on MambularRegressor) raises:

TypeError: mambular.models.utils.sklearn_parent.SklearnBase.fit() 
got multiple values for keyword argument 'regression'

The issue occurs because:

• SklearnBase.optimize_hparams() calls self.fit(..., regression=...)
• SklearnBaseClassifier.fit() and SklearnBaseRegressor.fit() do not declare a regression parameter in their signatures.
• They hardcode regression=False/True when calling super().fit(...) and also forward **trainer_kwargs.

This results in regression being passed twice to SklearnBase.fit(), causing the TypeError.

This makes optimize_hparams() unusable for both classifier and regressor models.

---

To Reproduce

from mambular.models import MambularClassifier

model = MambularClassifier(
    d_model=64,
    n_layers=4,
)

model.optimize_hparams(
    x_train_preprocessed,
    y_train,
)

Error:

TypeError: mambular.models.utils.sklearn_parent.SklearnBase.fit() 
got multiple values for keyword argument 'regression'

The same happens with MambularRegressor.

---

Expected behavior

optimize_hparams() should run without raising a TypeError.

regression should only be passed once to SklearnBase.fit().

---

Desktop (please complete the following information):

• OS: Ubuntu (local machine)
• Python version: 3.11
• Torch version: 2.5.1+cu121
• Mambular version: 1.5.0

---

Additional context

The problem appears to be a mismatch between SklearnBase.optimize_hparams() and the fit() signatures in SklearnBaseClassifier and SklearnBaseRegressor, leading to duplicated keyword arguments.

[DavidSolan0]

Proposed solution

Instead of hardcoding regression=False/True inside SklearnBaseClassifier.fit() and SklearnBaseRegressor.fit(), the method should explicitly accept a regression argument and forward it to super().fit().

For example:

def fit(..., regression=False, **trainer_kwargs):
    return super().fit(
        X=X,
        y=y,
        regression=regression,
        ...
        **trainer_kwargs,
    )

This avoids passing regression twice, and fixes the TypeError when calling optimize_hparams().

The same adjustment would apply to SklearnBaseRegressor.