Petar Zecevic

Results 10 comments of Petar Zecevic

Does the file have a lot of http and file links? In that case I think you can speed it up with the fix described in #1824.

@Wetlize Can you please elaborate why is it necessary for roam to store http, https, file and other links to its database? You said in #1824 that it's a feature...

OK, thanks. I think I can stick with only storing id links for now.

Yes, using kmeans initialization does help in 2D. But when I try it in 7D (the case I'm really interested in), I get the following error: ``` ~/anaconda3/envs/multiproc/lib/python3.8/site-packages/pygmmis.py in fit(gmm,...

`initFromKMeans` works OK it seems. However, when GMM object is initiated from KMeans, then for some reason after an M step the A2 array contains all `nan`s. `_Msums` seems to...

Peter, could you please take a look at this repo: [https://github.com/zecevicp/pygmmis_test](https://github.com/zecevicp/pygmmis_test) ? I placed a notebook there with a minimal example to reproduce the error and the accompanying data.

Unfortunately, cutoff of 3 or 5 produces the same exception (sometimes the message "Singular matrix" appears). Again, this only happens when KMeans clustering is the init method...

K-means initialization looks OK to me. But somehow it gets lost after a few iterations (for some reason the A2 array after the M step contains all NaNs). Default initialization...

I'll try to clarify... I'm interested in the 7D case. 2D case was to see how pygmmis takes care of the case where scikit-learn implementations fail. It handles it well,...

Same issue here with WSL 2.3.12.0