Keras Adding a predict_proba method on SKLearnClassifier

Currently the SKLearnClassifier only exposes a predict method. At the moment the raw probabilities are used in the predict method like:

# keras/keras/src/wrappers/sklearn_wrapper.py
class SKLBase(BaseEstimator):
    ...
    def predict(self, X):
        """Predict using the model."""
        from sklearn.utils.validation import check_is_fitted

        check_is_fitted(self)
        X = _validate_data(self, X, reset=False)
        raw_output = self.model_.predict(X) # <----- here
        return self._reverse_process_target(raw_output) # <----- transformed back

If a user wants to get probabilities out, they need to create a custom class like:

class MyKerasSKLearnClassifier(SKLearnClassifier):
    def predict_proba(self, X):
        if not hasattr(self, 'model_'):
            raise RuntimeError("You must fit the model before calling predict_proba.")
        return self.model_.predict(X) # returns the raw probas

Would you consider adding a predict_proba method to the SKLearnClassifier class:

class SKLearnClassifier(ClassifierMixin, SKLBase):
    ...
    def predict_proba(self, X):
        """Predict class probabilities of the input samples X."""
        from sklearn.utils.validation import check_is_fitted

        check_is_fitted(self)
        X = _validate_data(self, X, reset=False)
        return self.model_.predict(X)

so that downstream implementations and users can rely on the default class imported from Keras, rather then creating a custom wrapping class. predict_proba is the common method in the scikit-learn library for getting the probabilities out of a classifier.

Comment From: divakaivan

If ok, I can open a PR :)