is instability increased with random row ordered? - Githubissues

ai-se / Pits_lda

IST journal 2017: Tuning LDA

https://github.com/amritbhanu/LDADE-package

4 stars 4 forks source link

is instability increased with random row ordered? #6

Closed amritbhanu closed 7 years ago

amritbhanu commented 8 years ago

[x] Pitsabcde
[x] wikipedia

amritbhanu commented 8 years ago

New Results

Just Pits A dataset
Same row order
Used Scikit learns, LDA package, to find the relevant topics. Multiple runs, didnt give that many instable topics.
Only in the implementation of spark mllib lib, it is very unstable.
Both implementation has the basic steps like online learning, gibbs sampling.
Just some parameters are different, this might be the reason for such a huge difference.

file

random row ordered results

Implemented using scikit-learn lda
Topics remained unstabled across multiple runs for pits A
Ordering to top topics are changed.

file