jasperyang
The algorithm is very slow
Thanks for the contribution. I am testing the algorithm with 120K of tweets, but it is extremely slow. 25 hours passed so far and it has not started to the iterations yet! Do I have any mistake? I read from the main paper that BTM is nearly 3 times slower than LDA, but according to my experiments, I think it is at least 10imes slower than LDA Gensim implementation. What do you think about that? Is there any approach providing faster processing?
That is too slow, i haven't test the speed. But my code is finished by numpy array, so in python it maybe is the best speed already. You can try to write it in c++ which helps a little in speed. In industry, people usually develop a distributed beta of LDA , which is totally a F1 racer.
