Recently, I found out that the Sklearn NMF library deprecated its pg solver and now uses its cd solver. With the cd solver, I don't think I can apply sparseness constraints. Or maybe the L1 rate implicitly represents sparseness? Is there a way to apply sparseness constraints?
Because I want to make the solution space smaller in NMF, I want to specify sparseness constraints to both improve performance and decrease solution space size. Is there an alternative to specifying sparseness constraints?
