Suppose I create EM object, train it and then change its "nclusters" property. In my understanding, this should reset it to the untrained state, since the sizes of means and covs do not satisfy the number of clusters.
However, isTrained() returns true.
Is is a bug or feature?