Warning when split or fold does not contain at least 1 of each class

[aiqc]

Background

Splitsets and foldsets divide the sample population into evenly distributed sets. However, in unbalanced datasets or highly folded datasets you can't always guarantee that infrequent classes will be present in each split/fold.

Problem

If a split or fold does not contain at least 1 of each class, it can cause downstream problems during encoding and when producing metrics. AIQC attempts to prevent these scenarios by encoding on all samples.

Solution

However, it would be good to warn the user that their splits/folds are not representative of the larger population. For Splitsets with a categorical label, warn the user at the end of Splitset or Foldset creation if any of their splits do not contains at least 1 of each class.