QUESTION - low cpu usage with Parallel() prefer='threading'

[Blnjjj]

Hi all,
Thank you for providing such a wonderful tool.

I have some issue while using the TimeSeriesKmeans in tslearn.
I found that in the fit() function, even I set n_jobs = -1 the cup usage on both my Mac and linux machine remains to be low.

So I tried to make some change on the _cdist_generic() function, commit out the prefer='threading' and add backend='loky', and it seems to speed up the clustering.

I'm not so familair with this so I don't know why threading is not working on my side and whether my change will bring trouble to the fitting result.

Any help to this is grateful! Thanks in advance.