[scikit-learn] New member of the triage team: Norbert

Olivier Grisel olivier.grisel at ensta.org
Mon Jun 21 11:11:33 EDT 2021


> I have only one question related to scikit-learn.
> how to compute topic coherence of lda models in scikit-lean.  I don't find any function that calculate a coherence value.
> please, reply me.

We don't have such a metric in scikit-learn. I assume you are referring to:
http://svn.aksw.org/papers/2015/WSDM_Topic_Evaluation/public.pdf

which is implemented in Gensim as:
https://radimrehurek.com/gensim/models/coherencemodel.html

If I understand correctly this metric needs to compute relative
frequencies of occurrences and co-occurrences of words in the
documents of the training set. This feels very domain specific
compared to the more domain agnostic metrics that we have in
scikit-learn.


More information about the scikit-learn mailing list