markopy

I'm also here because importing umap (https://github.com/lmcinnes/umap/issues/631) / pynndescent (https://github.com/lmcinnes/pynndescent/issues/111) causes a ~10 second delay which makes it impossible to use these modules in a command line application, for example....

Interesting, are you suggesting that using `parallel` might somehow disable caching? I did a profile of the pynndescent issue and the main delay seems to be in numba's `compiler_machinery.py`.

+1 for this. It makes sense to just check out the branch actually needed and without all the history. This can reduce the cloning time enormously for large repositories with...