More (julia) packages to interface with

[tlienart]

This is a list to keep track of packages that should be considered in the near future with some notion of priority; feel free to comment with packages that should be considered or shotgunning an interface you'd be happy to help with.

higher priority

• [x] Complete interface with DecisionTree.jl e.g. categorical features part
• [ ] Impute.jl
• [x] Soss.jl at least for basic bayesian models such as bayesian regression or bayesian logreg --> this is happening on branch soss
• [x] Finish the interface with Multivariate Stats (e.g. subspace LDA is missing)
• [ ] Complete the interface to Clustering.jl

other models

• [ ] LowRankModels.jl and also this mlj issue
• [ ] ManifoldLearning.jl also this mlj issue
• [x] EvoTrees.jl and this mlj issue~~
• [ ] tsne
• [ ] umap
• [ ] NearestNeighborDescent

---

• [ ] https://github.com/JuliaStats/StatsKit.jl

• [ ] https://github.com/Marco-Congedo/PosDefManifoldML.jl note: objects are matrices, could maybe use the images formalism.

[ablaom]

Maybe add to this list unimplemented parts of DecisionTree package, eg the categorical features part.

[tlienart]

• https://github.com/TensorDecompositions/NTFk.jl (Nonnegative Tensor Factorization + k-means clustering) but GNU licensing
• https://github.com/TensorDecompositions/NMFk.jl (Nonnegative Matrix Factorization) also GPL

[tlienart]

https://github.com/theogf/AugmentedGaussianProcesses.jl

https://github.com/juliangehring/Bootstrap.jl

wrap blingfire

[tlienart]

@OkonSamuel I realize QDA is nowhere to be seen in Julia, maybe that's something you'd be interested in taking up? there's this old package: https://github.com/trthatcher/DiscriminantAnalysis.jl which offers it but I reckon it'd be a fine thing to add to MultivariateStats or have as a standalone model 🤷‍♂ ?

[OkonSamuel]

@tlienart I'd love to add a QDA model implementation . But https://github.com/trthatcher/DiscriminantAnalysis.jl package seems old (especially julia syntax and standard libraries used) and might not be be compatible with julia 1.0+. A quick look at the compat section in the project.toml file shows this. Although the dev branch of the repo fixes this and some other compatibility problems.

[tlienart]

Yes sorry I should have made this clearer: you can either

• extract the QDA code and PR it to MultivariateStats
• fork the DiscriminantAnalysis package and make it up to date

I think the first option is probably best given that you're already familiar with the MV package, but if there's a lot of code to port then maybe the ssecond option is better

On our side if there's QDA it'd be nice (though there soon will be via sklearn)

[tlienart]

https://github.com/davidavdav/HClust.jl though would require to basically start anew given it's 4 years unmaintained

[jbrea]

Would be nice to have UMAP, T-SNE and maybe other dimension reduction methods like Phate in MLJ. @ablaom: where should this be done? In an existing package or a new package. e.g. MLJDimensionReduction.jl?

[ablaom]

I see UMAP has a julia interface And there is https://github.com/lejon/TSne.jl, but that's looking pretty old. I'm not sure about Phate. Were you offering to provide direct interfaces to python (ok) or C (better) libraries yourself? These could go in new standalone packages, or you could collect them in a single package if you thought that made sense. Maybe call it DimensionReduction.jl and JuliaAI could host it.