We may derive some variants from the proposed KNN version. We may assign different weights to selected neighbors instead of identical ones: the highest weights to the first nearest neighbor and the lowest weight to the last one. Instead of a fixed number of nearest neighbors, we select any number of training examples within a hyper-sphere whose center is the given novice example as neighbors. The categorical scores are computed proportionally to similarities with training examples, instead of selecting nearest neighbors. We may also consider the variants where more than two variants are combined with each other.