[Topic-models] negative values among LDA's thetas?

Lara Michaels laramichaels1978 at yahoo.com
Fri Jul 23 14:16:43 EDT 2010

Dear all,

I am again running David's LDA C code on a different dataset and was following the helpful advice I was given earlier


on how to compute the probabilities of each document belonging to the different topics. However, I just noticed that with this second dataset some lines of the resulting theta file have multiple negative values for the probabilities and then a value larger than 1. Eg, the first line from the thetas file (using a model with 14 topics):

-0.000104890502391142 -0.000104890502391142 -0.000104890502391142 -0.000104890502391142 -0.000104890502391142 1.00136357122330 -0.000104885194608265 -0.000104890502391142 -0.000104890502391142 -0.000104890502391142 -0.000104890502391142 -0.000104890502391142 -0.000104890502391142 -0.000104890502391142

Is this a sign that I messed up at some point? Or just that this document very much belongs to the sixth topic? : )

many thanks for any insight,


More information about the Topic-models mailing list