RFC: Allow AnomalyDetector models to act as transformers.

[milesgranger]

Topic

#368 created the concept of having AnomalyDetector type models which add an anomaly functionality to any base_estimator

Potential Feature

Implement, in some way, that this can be used inside of a Pipeline as a Transformer. So that perhaps one could use such a model in a FeatureUnion to further predict an anomaly via another type of AnomalyDetector

Probable Implementation?

One possible solution is to implement .transform method on the AnomalyDetector model, such that all training would occur there with base_estimator and then output the anomaly calculation(s) as features.

[milesgranger]

I think this is so 'far out' from being used by any DS, that we might as well close it until it comes up in a more realistic scenario.