OutlierMixin

[mahynski]

SIMCA_Model and EllipticManifold_Model are possibly better classed as outlier detectors so should inherit OutlierMixIn from sklearn not Classifier mixins. Consider changing this. The *_Authenticator versions can be kept as classifiers since that is more consistent with how the end user will apply them.

[mahynski]

Outlier mixin convention is to set -1 = outlier vs +1 as inlier during prediction (https://scikit-learn.org/dev/modules/generated/sklearn.base.OutlierMixin.html#sklearn.base.OutlierMixin). I prefer the True/False boolean convention since it makes other steps, like screening outliers etc. simpler and more sensible. In addition, outlier detectors are also distinguished from novelty detectors in sklearn; this convention is not always clearly transferrable to many of our applications so for now, we will leave these as "classifiers" not "outlier detectors".