sklearn tests

[wcwitt]

Hi @cortner, I see you disabled some sklearn tests recently, possibly because they were failing with the new backend?

ACEpotentials.jl/test/test_silicon.jl

Line 104 in 15d12b8

# @testset "SKLEARN_ARD" begin

Noam (with my help) is trying to reproduce some past fits and we need this functionality ... would it be best to work from an older ACEpotentials?

[wcwitt]

Here's the commit that disabled them: f6d42d3.

[cortner]

they have been failing for a long time. It has nothing to do with the new the backend, I actually haven't retired the old backend yet.

I disabled them because I wanted to rethink carefully about how to re-integrate SKLEARN parameter estimation. Likely through an extension or through some other generic Julia linear regression packages.

[wcwitt]

That's helpful - maybe it has to do with the MLJ extension. I'll try to figure it out.

[wcwitt]

Tests passed here: https://github.com/ACEsuit/ACEpotentials.jl/actions/runs/9090697108

[wcwitt]

This issue will be partly addressed by ACEsuit/ACEfit.jl#77.

I disabled them because I wanted to rethink carefully about how to re-integrate SKLEARN parameter estimation. Likely through an extension or through some other generic Julia linear regression packages.

Happy to discuss this, but at the moment sklearn is available only through extensions. Although it's a little confusing because there are two routes: one utilizing PythonCall as an extension and the other through MLJ as an extension.

[cortner]

If thought about this and I'd like to avoid installing all the python dependencies for the ACEpotentials tests so would prefer not to re-instate those. If you disagree with me please re-open and we can discuss again.