Negative log_perplexity

[Jonathan Peters]

Hello,

I created an LDA model from control data and I am trying to calculate the perplexity of my test data. I am using the ldamodel.log_perplexity function to do that.

Example: perplexity = lda_model.log_perplexity(test_corpus)

I am getting a negative output (-48.51) and I am wondering what exactly that means since in papers I've read, their output is positive.

Would be great if you can provide a source as well!