lucidrains / audiolm-pytorch

Implementation of AudioLM, a SOTA Language Modeling Approach to Audio Generation out of Google Research, in Pytorch
MIT License
2.32k stars 249 forks source link

Dependency error #226

Closed amrzv closed 10 months ago

amrzv commented 10 months ago

Hi. When running HubertWithKmeans the Assertion error is raised because scikit-learn doesn't match required version 0.24.0

image

Looks like scikit-learn==0.24.0 should be specified in setup.py

P.S: tested in colab. P.S.S: current version in colab is scikit-learn==1.2.2.

lucidrains commented 10 months ago

@amrzv i ran into some weird error last time and was too lazy to figure it out

could you just try running the command as per the assert instructions?

lucidrains commented 10 months ago

iirc, i'm not even using scikit learn anymore. the cluster centers are extracted here

lucidrains commented 10 months ago

@amrzv yea let me know if it works with the manual pip install

we can leave this issue open, as i'd like to just suppress the warning in the fairseq library, and just extract the cluster centers and use them. if anyone is training their own set of cluster centers, they can deal with that scikit-learn version problem

lucidrains commented 10 months ago

removed the assert altogether