Add the possibility to declare a framework as abstract

[sebhrusen]

This is mainly useful to define a "template" framework from which children could extend to inherit all its properties without wanting to be able to run the template itself.

A typical use-case is MLPLan that comes with 2 training engines.

[Innixma]

Would it also make sense long-term to convert the def run(dataset, config): to some kind of class FrameworkRunner?

A lot of the data loading code is shared between all of the frameworks, and they can be viewed as individual ML models that take data in and generate output aka output = model.fit(data, **kwargs). This would simplify the work for frameworks to have slightly modified variants of their exec.py, and would allow for an AbstractFrameworkRunner.

[sebhrusen]

@Innixma That's a good idea.
I agree that the run.py is too monolithic currently.
A class would allow to structure this and improve code reusability, sth like:

class FrameworkRunner:
    def __init__(self, config, dataset): pass
    def prepare_data(self): pass
    def fit(self, …): pass
    def predict(self, …): pass
    def get_result(self): pass

Still need to figure out the right API, but it probably makes sense as a future improvement.

[sebhrusen]

#279