Closed marinadudarenko closed 9 years ago
When using large tau coeffitients in sparsing DirichletPhi or DecorrelatorPhi regularizers, final document-topic distributions is equal to zero. Possible problem may be in nulling word-topic raws in phi matrix.
Let's also have a Score to count how many situations like this ocured during iteration
Re-open as https://github.com/bigartm/bigartm/issues/91
When using large tau coeffitients in sparsing DirichletPhi or DecorrelatorPhi regularizers, final document-topic distributions is equal to zero. Possible problem may be in nulling word-topic raws in phi matrix.