Add MinimalFeatureExtractionSettings about tsfresh

[nils-braun]

Do you think it would be better to delete the other calculations from the beginning in the name_to_param array or just do not add them to the kind_to_calculation_settings_mapping in the set_default_parameters function?

And how to deal with it in the from_columns method? I think here the functions should still be added if the user requests them - this is why I would not delete them from name_to_param but from kind_to_calculation_settings_mapping.

[MaxBenChrist]

@jneuff looked into that. Maybe he can add some remarks as well. We moved the setup of name_to_param from the set_default_parameters to the constructor. See 8048ef4.

The current logic is "extract all the features from name_to_param for all types of time series".

If we follow that for a minimal MinimalFeatureExtractionSettings Object, then it just have a smaller name_to_param object.

[jneuff]

I'd offer methods to instantiate a minimal and a full Settings object. A simple solution would be to initialize name_to_param as empty dict in the init and provide two class methods that will instantiate and configure a minimal and a full Settings object respectively. Similar to what we're doing right now with from_columns.

[nils-braun]

I am actually against doing this with another member function - because it would be something everyone has to call everytime and it would be easily forgotten. I would rather create a child class MinimalFeatureExtractionSettings and overload the constructor.

I can implement this and we can talk about it again maybe.