Calculation of <prob_topics> in <Latent Dirichlet Allocation.ipynb>

[DaveRockt]

Hello,

is the calculation of the conditional probability of assigning each topic <prob_topics> correct?
It does not seem to be the same as in the referenced Wikipedia article.

Best,
David

[hammadshaikhha]

@DaveRockt Thanks for raising the issue, I just noticed this. Could you be a bit more specific and compare the formula I have with the one in wikipedia?

I will review my LDA notebook in the meantime and see if I can spot a mistake.

[DaveRockt]

Thank you for your answer and sorry for not being specific:

In the formula on Wikipedia, I cannot find what you call 'denom1'. In the meantime, however, I found out that after normalising, I get the same result as you.

However, I have another question:

In your code in the 'Main part of LDA algorithm' under 'Add in current word back into count matrixes', you use 'init_topic_assign'. Does this make sense? Shouldn't you use the new assigned topic?

Thank you for this project by the way, it really helped me to understand LDA better.

Best,
David

[hammadshaikhha]

Hi David,

Sorry for the late reply, I skimmed over the code and you may be right. I think one way to check would be to use a python library and run LDA on the same data set and see whether the results match. Could you work on doing that?

If there is indeed a mistake in the code, feel free to fix it and do a pull request.

[DaveRockt]

Thank you. At the moment I am also a little bit busy. But I will have a look asap. Best, David Von: Hammad Shaikh [mailto:[email protected]] Gesendet: Montag, 16. Juli 2018 21:11 An: hammadshaikhha/Math-of-Machine-Learning-Course-by-Siraj <[email protected]> Cc: Bälz, David (ECON) <[email protected]>; Mention <[email protected]> Betreff: Re: [hammadshaikhha/Math-of-Machine-Learning-Course-by-Siraj] Calculation of <prob_topics> in <Latent Dirichlet Allocation.ipynb> (#2) Hi David, Sorry for the late reply, I skimmed over the code and you may be right. I think one way to check would be to use a python library and run LDA on the same data set and see whether the results match. Could you work on doing that? If there is indeed a mistake in the code, feel free to fix it and do a pull request. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#2 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/Amz8fubvCBTTcGWc5s4ge38pHiMuND1wks5uHOVZgaJpZM4U9Mo6>.