[gensim:11592] Seeded LDA | guided lda | using eta to guilde LDA

[r***@qedata.io]

Hi There, I have been trying to use eta to create a topic word matrix with
values to guide the LDA process.

My expectation is the the LDA will converge using the set eta and my topics
and topic order will be known.

For example: if I seed:
[0][word_index] then know 0 is topic 1
[1][word_index] then know 0 is topic 2
[2][word_index] then know 0 is topic 3
[3][word_index] then know 0 is topic 4

When I take the top 10 words for each topic the order is all scrambled.
They seemed to be correctly clustered but the order is all off.

topic 1 is at index 3
topic 2 is at index 0
etc..


Am I setting eta wrong?

In the end I want to use gensim like this algorithm.
https://github.com/vi3k6i5/GuidedLDA/blob/master/guidedlda/guidedlda.py

Please help

[r***@qedata.io]

Correction:

For example: if I seed:
[0][word_index] then index 0 is topic 1
[1][word_index] then index 1 is topic 2
[2][word_index] then index 2 is topic 3
[3][word_index] then index 3 is topic 4

Post by r***@qedata.io
Hi There, I have been trying to use eta to create a topic word matrix with
values to guide the LDA process.
My expectation is the the LDA will converge using the set eta and my
topics and topic order will be known.
[0][word_index] then know 0 is topic 1
[1][word_index] then know 0 is topic 2
[2][word_index] then know 0 is topic 3
[3][word_index] then know 0 is topic 4
When I take the top 10 words for each topic the order is all scrambled.
They seemed to be correctly clustered but the order is all off.
topic 1 is at index 3
topic 2 is at index 0
etc..
Am I setting eta wrong?
In the end I want to use gensim like this algorithm.
https://github.com/vi3k6i5/GuidedLDA/blob/master/guidedlda/guidedlda.py
Please help

[r***@qedata.io]

Solution was that in PyLDAvis there was shuffling of the return dataframe.

Other problems of eta exist. Specifically the an eta between 0->1 does not
provide help and requires a multiplicative factor to ensure expected
results.

Post by r***@qedata.io
Hi There, I have been trying to use eta to create a topic word matrix with
values to guide the LDA process.
My expectation is the the LDA will converge using the set eta and my
topics and topic order will be known.
[0][word_index] then know 0 is topic 1
[1][word_index] then know 0 is topic 2
[2][word_index] then know 0 is topic 3
[3][word_index] then know 0 is topic 4
When I take the top 10 words for each topic the order is all scrambled.
They seemed to be correctly clustered but the order is all off.
topic 1 is at index 3
topic 2 is at index 0
etc..
Am I setting eta wrong?
In the end I want to use gensim like this algorithm.
https://github.com/vi3k6i5/GuidedLDA/blob/master/guidedlda/guidedlda.py
Please help