num_clusters now set to the minimum between requested number and the number of features in the input dataset