Regularization strength for logistic (regression) classifier

[lars-reimann]

Is your feature request related to a problem?

There should be a way to control regularization of the LogisticClassifier.

Desired solution

Add an optional, keyword-only constructor parameter c: float = 1.0 and pass it to the wrapped scikit-learn estimator.

Possible alternatives (optional)

I've originally also considered letting the user choose the penalty ("l1"/"l2"/"elasticnet"). However, most solvers only support "l2" anyway. We should rather stick to "l2" and later choose an appropriate solver internally based on the shape of the data for fast convergence (different issue).

Screenshots (optional)

No response

Additional Context (optional)

No response