pynndescent 0.6.0 bug
Author: timothee-lCreated Feb 9, 2026Updated Feb 9, 2026
It looks like there is an issue with pynndescent v0.6.0 that yields unexpected results.
v0.5.13
v0.6.0
In this dataset, the right most cluster you see on the first plot contains very similar embeddings (in the original space) but in the second plot, these points are instead scattered across the whole plot.
I'm running python 3.13.5 on apple silicon, umap-learn==0.5.11 numpy==2.1.3 numba==0.63.1 scipy==1.17.0 scikit-learn==1.8.0 for both figures.
Source: lmcinnes/umap