python - Perplexity comparision issue in SKlearn LDA vs Gensim LDA -

i applied lda both sklearn , gensim. checked perplexity of held-out data.

i getting negetive values perplexity of gensim , positive values of perpleixy sklearn. how compare values.

sklearn perplexity = 417185.466838

gensim perplexity = -9212485.38144

in order compare perplexities need convert gensim's perplexity using np.exp(-1. * gensim_model.log_perplexity(train_corpus)).

see here general comparison of gensim , sklearn lda implementations.