Add an option for users to use pyspark to parallelize the training of multiple NMFs for each hyper-parameters.