cbc385ca10
save training dataset by it target type. while training NN use single target training set to backward NN. this improve at least 20 times faster than last update!