The `python` and `scipy` version-compatibility, and KLDivergence() needs attention!
prikarsartam opened this issue · comments
- OCTIS version: 1.13.1
- Python version: 3.12.0
- Operating System: Linux-64
I am not an expert in package management so I do not fully understand all the details of it. octis
installs properly in google colab, but installing in kaggle requires pip install octis --use-pep517
.
Now installing locally on my system I had the following issue - both for installing with pip install octis
and pip install -e.
from the downloaded repository which is of prior concern to me.
Description
- Installing with the latest python3.12 in my linux doesn't end successfully in any case as
zipimport
has been deprecated from Python3.10 onwards. - Since this repo requires gensim==4.2.0 it has
insidegensim/matutils.py
but to the best of my knowledge thetriu
has been deprecated for scipy==1.13.0 onwards. - Also the
KLDivergence
inoctis.evaluation_metrics.diversity_metrics
returnsRuntimeWarning: invalid value encountered in log divergence = np.sum(P*np.log(P[/Q](http://localhost:8888/Q)))
What I Did
I made a conda virtual environment with python3.10 and downgraded scipy==1.12 : so prob 1 and 2 are solved.
For the case of 3 : the model_output['topic-word-matrix]
for ProdLDA
is not suitably normalized in [0,1]
to be interpreted as probabilities which gives negative entries in the matrix leading to nan
in np.log()
.