Multivariate classification support #74
Description
Multivariate classification in GW context is a tricky issue due to spatial heterogeneity of presence of individual classes. Some local models might be predicting all 8 classes while other 4, or 2. That makes the evaluation of the entire model wildly complicated as accuracies mean different things and so on. Yet, there might be situations where the model is possible to be fitted (balanced class presence in local neighbourhoods), so we should be able to support it. At the moment, classifiers are artificially limited to binary y so this shall be relaxed and some additional tooling needs to be adjusted.
Note that this is not an easy issue to start with.