Request to add LaplaceRedux.jl to the MLJ registry

[pasq-cat]

we at Taija would like to add LaplaceRedux.jl to the MLJ's model registry.
we have written an interface for MLJ laplaceredux.jl interface,

LaplaceRedux.jl allows to compute the approximated posterior predictive distribution of any model through the Laplace approximation (as explained inhttps://arxiv.org/abs/2106.14806) and grants the users the possibility to compute credible intervals with a low computational overhead.
the LaplaceRedux.jl package is part of the taija (Trustworthy Artificial Intelligence in Julia) effort (https://github.com/JuliaTrustworthyAI) effort.

[ablaom]

This looks very interesting, thanks. Just back from leave so it may take a week or two before I can look at this properly.

Note to self: Uses some non-public API from MLJFlux.

[ablaom]

cc @EssamWissam

[ablaom]

In the meantime @pasq-cat you may want to upgrade your doc-string to make it MLJ compliant.

[pasq-cat]

@ablaom here https://github.com/JuliaTrustworthyAI/LaplaceRedux.jl/blob/main/src/direct_mlj.jl thank you for your time

[ablaom]

Thanks for this. I can hopefully take another look in the next two weeks. Thanks for your patience.

[ablaom]

To do list for @ablaom:

• Check level=4 integration tests pass
• Check document strings are compliant
• Doc-string issues raised below have been addressed
• Add to the MLJModel registry
• Add model descriptors for MLJ Model Browser and generate new MLJ docs
• Add to MLJ integration tests

[ablaom]

@pasq-cat According to the docs for the regressor:

• X: any table of input features (eg, a DataFrame) whose columns
  each have one of the following element scitypes: Continuous,
  Count, or <:OrderedFactor; check column scitypes with schema(X)

However, the input_scitype makes no mention of Count. Is this a mistake in the docstring or the input_scitype declaration?

[ablaom]

Can I suggest avoiding unicode subscript characters for hyper-parameter names, as in μ₀. So μ0 or mu0 are friendlier. Of course I realize this breaking, unless you provide both options, like this mu0 = 0.0, μ₀=mu0 (but do use mu0 in the documentation).

[ablaom]

Okay, I think the issues are now sorted on master. Can you please tag a new patch release. Then I can go ahead with registration.

[pasq-cat]

Okay, I think the issues are now sorted on master. Can you please tag a new patch release. Then I can go ahead with registration.

@pat-alt can you do it? I am working this week and I come home really late.

[pat-alt]

Thanks @pasq-cat. I've fixed the remaining things in JuliaTrustworthyAI/LaplaceRedux.jl#135, merged the PR and published a new release: https://github.com/JuliaTrustworthyAI/LaplaceRedux.jl/releases/tag/v1.2.0

@ablaom hopefully this should do this trick? Thanks once again for all your help with this and do let us know if anything else needs to be addressed.

[ablaom]

Thanks once again for all your help

Your welcome.

There's another model being registered soon, so there may be a small delay - it's less work to do both together.

[ablaom]

Thanks for your patience. There is a hiccup at MLJModels with the new version of Julia - something to do with Statistics compats (Julia "bug"). It looks like it will be sorted shortly and then I'll go ahead with your model.

[ablaom]

Closed as complete.

[ablaom]

Closed as complete.