Closed consideRatio closed 2 years ago
@yuvipanda do you have intuition if removing nomkl
as was installed in the base image (pangeo/pangeo-notebook) will cause issues for us with performance or similar?
I think this is totally fine, thanks for digging in, @consideRatio!!
I can't vouch for the reasoning behind how the pangeo image was built and why they chose to include the nomkl
package, but it's completely fine for us to not do that.
They may have been concerned about MKL itself not being open source or its size on disk, but at our usage patterns neither of those is a blocker.
Note: obviously I'd prefer if MKL were open source, but this isn't the place to worry about that. And since it is used as a very low-level optimization by numpy, with a--less performing but functional--viable alternative in OpenBLAS, it's fine to use it in our work. While not open source, it is freely available in conda-forge for use with all numpy/etc installations, so we have no concerns about its use.
The workaround of removing
nomkl
seem to enable a successful installation of pymc3. I thankfully couldn't see a reinstallation ofnumpy
orpandas
which would be a sign of a problem. Maybe removingnomkl
is okay?