Jeremie Desgagne-Bouchard

@ablaom If you have time, let me know what you think of https://evovest.github.io/EvoTrees.jl/dev/. Models in particular are found under the [Models](https://evovest.github.io/EvoTrees.jl/dev/models/) section. Working out the docs, I had a few...

Very impressive performance! I thought I had managed to get decent optimisation on the histogram building and that remaining speed bottle neck was mainly coming from the handling for the...

@brianhempel Thanks a lot for providing those explanations. I've started makig some refactoring in the `dev-refactor` branch. The optimizations per haven't been incorporated, but I've looked to adapt the structure...

Thank you for these hints! I've spent most of my recent efforts to bring the EvoTrees ready for a release baesd on the restructuring to better handle the various optimisations....

Hi! Thanks for bringing this suggestion. I wasn't aware of that Cubist package. It seems to be similar to what is perfomedby piecewise-linear [library](https://github.com/GBDT-PL/GBDT-PL). There are some aspects of the...

I can see that such approach for a linear predictor would be functional. For what I can see, it would however be a costly operator as evaluation of the split...

EvoTrees is effectively designed around dense matrix, so the poor performance on sparse matrix is unfortunately expected. The library was initially built around my uses cases which didn't involve sparse...

Sure, here are 2 docs about it with CatBoost: https://arxiv.org/pdf/1706.09516.pdf http://learningsys.org/nips17/assets/papers/paper_11.pdf For lightgbm: https://lightgbm.readthedocs.io/en/latest/Features.html#optimal-split-for-categorical-features, though I haven't access to the referenced 1958 paper :) There are a couple of ways...

Effectively, `fit` expects that data provided has went through the `reformat` step. However, `MLJBase.fit!` works fine on tabular data, and you can actually start training using that function as well,...

I'm assuming this can be closed. Feel fre to reopen otherwise.