Create general tool for all the use cases

[perellonieto]

Most of the actual tools have several things in common and could potentially be merged into one parent tool.

• All of them expect a model for the initialisation
• All of them expect input matrices with x_tr, or x_te, y_tr, or y_te
• All of them expect the passed model to have a function to fit or partial_fit
• All of them expect the model to generate some outputs like scores, or features

It could be possible to specify what functions to call during training and a dictionary to return the output values.

[tdiethe]

Sounds quite nice and flexible. You could probably make use of the scikit-learn base classes to do type checking (or just do duck-typing), see e.g. here

[perellonieto]

That looks great, Thanks. I think I should use these BaseEstimators as a standard and create tools for each particular type.