mdabros / SharpLearning

Machine learning for C# .Net
MIT License
383 stars 84 forks source link

XGBoost lib is dead #156

Open madskonradsen opened 4 months ago

madskonradsen commented 4 months ago

Hej Mads,

It looks like the PicNet XGBoost lib is no longer maintained (and is still stuck on an ancient dotnet version) as per 4 months ago: https://github.com/PicNet/XGBoost.Net/commit/28902d9a9dddb3bd09fdcc42aafa191e96f10f3e The lib was of course a bit outdated, and now that XGBoost 2.0 is out, it's even more outdated, which is a shame since it's a nice library :) I've tried looking for viable alternatives to the PicNet package, but it doesn't look like there's anything out there.

So the question here is more of: should XGBoost be removed from the SharpLearning portfolio, or are you aware of any libs that might be able to fill the spot?

mdabros commented 3 months ago

Hej Mads ;-).

I am actually working on a replacement for XGBoost.NET based on XGBoost 2.0. It is based on the code from XGBoost.NET, but will eventually be more feature complete. I need to fix a few practical things related to CI and Linux support before I publish the code and package.

Once released, I will use this from SharpLearning moving forward.

madskonradsen commented 3 months ago

Nice! :) In case you want help, consider creating a few tickets in the issue tracker :)

mdabros commented 3 months ago

Will do once published. You are more than welcome to help! :)

MiWeiss commented 1 week ago

Hey ~@madskonradsen~ @mdabros

Do you have any update on the progess regarding theXGBoost.NET replacement? I've been considering writing one myself, but seeing that you're already working on that, I might be better off just waiting for a bit 😄

madskonradsen commented 1 week ago

I think you tagged the wrong Mads :) @mdabros @MiWeiss

mdabros commented 1 week ago

Hi @MiWeiss

Yes, it is more or less ready. For now, same functionality as the old XGBoost lib, but with code clean up, targeting dotnet 8, and native packages for windows, linux, and osx for xgboost 2.0. I will see if I can publish it in this state during the next week. Then we can work on adding in new features from there. You are of course more than welcome to contribute once it is published.

mdabros commented 21 hours ago

Hi @MiWeiss and @madskonradsen

I have release the renewed version of the old XGBoost.Net wrapper. The new name is XGBoostSharp (following the trend of TorchSharp). The first release is on nuget. Please try it out, and as mentioned, you are very much welcome to contribute. See more here: https://github.com/mdabros/XGBoostSharp

Initial release notes here: https://github.com/mdabros/XGBoostSharp/releases

best regards Mads

MiWeiss commented 21 hours ago

Thanks a lot @mdabros.