RuntimeError when printing step made from BaseEstimator subclass

[jrderuiter]

What is the bug?

When printing a representation of a step made from a BaseEstimator subclass (see below), I get the following runtime error:

RuntimeError: scikit-learn estimators should always specify their parameters in the signature of their __init__ (no varargs). <super: <class '_StepBase'>, <SklearnTransformer object>> with constructor (*args, **kwargs) doesn't  follow this convention.

How to reproduce it?

from baikal import make_step
from sklearn.base import TransformerMixin, BaseEstimator

class MyTransformer(TransformerMixin, BaseEstimator):
    pass

MyTransformerStep = make_step(MyTransformer)
MyTransformerStep()

What behavior did you expect?

To not get a runtime error.

What versions are you using?

• baikal version 0.3.0
• scikit-learn version 0.22.1
• Python version 3.8.1

Any additional information? Add here any other information about the problem that could be relevant.

[alegonz]

Thank you for the bug report!

I could reproduce it here, too. It fails when the class you're making a step from lacks an __init__ method and resolves to a delegate of object.__init__ which has a (*args, **kwargs) signature which is incompatible with scikit-learn. I'll work on a fix which I plan to release as version 0.3.1. In the meantime, to make it work please define an __init__ method in your base class, if possible.

[alegonz]

@jrderuiter Just released v0.3.1 with that bug fixed.

[jrderuiter]

Great, thanks!