You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In the efficient lda example the phis for example are set to zero and reestimated in the export method of the SparseLDAInferencer. However, massTotal of the phis is not set back to 0 and so the Proportions get messed up. I worked around that by decrementing each mass manually to zero before the export call. like that the masstotal also gets decreased to zero.
The text was updated successfully, but these errors were encountered:
I'm seeing the effect of this as well in release 1.0, although I'm not sure failing to reset the massTotal is the problem. The SparseLDAInferencerexport() method begins with phis.foreach(_.value.zero()), which as I read the code (albeit not really knowing Scala) should be reseting the masses. However, dumping the total mass of the phis shows that it increases with each export(). There's definitely a bug in there somewhere.
In the efficient lda example the phis for example are set to zero and reestimated in the export method of the SparseLDAInferencer. However, massTotal of the phis is not set back to 0 and so the Proportions get messed up. I worked around that by decrementing each mass manually to zero before the export call. like that the masstotal also gets decreased to zero.
The text was updated successfully, but these errors were encountered: