Clusters can be found within a training dataset (transductive) then used to train a classifier, so as to extend the clustering model to new instances. See snippet at #4543 (comment)

@selwyth is apparently working on this.