Consider deferring to scikit-learn defaults

[SRFU-NN]

Instead of explicitly having the long list of input parameters in the init() method for classes that only has very small changes from scikit-learn, we should consider only defining the parameters we use, and let the rest be up to scitkit-learn. Example ( for ProcessOptimizer\learning\forest.py ExtraTreesRegressor):

    def __init__(self, min_variance=0.0, **kwargs):
        self.min_variance = min_variance
        super(ExtraTreesRegressor, self).__init__(**kwargs)

That way, we defer to scikit-learn for defaults for settings we do not set, while allowing the user to set whatever.

We lose the nice hints as to what can be set, but I feel like the user who wants to set them can should read the scikit-learn documentation anyway.

It would make our code clearer, and avoid errors when scitkit-learn updates their defaults.

Another downside is that we could be unaware of an update in defaults from scikit-learn, which could make brownie bee and bleeding edge behave differently.

[SRFU-NN]

SQBL and I discussed it.

The issue has cropped up twice in the last year, and it takes approximately 2-3 days to fix each time, all inclusive. This means that we must expect to us about a man-week per year on it.

We also risk changes in the default behaviour of ProcessOptimizer, which could cause tests to fail. But this would be a much less transparent failure mode, so it would probably take longer to fix.

Overall, it is probably not worth it to spend time on it at the moment.