Yann Dubois

@RAMitchell I think this would be great. Currently, there are some new features that are not usable due to using the old binary format, for example categorical features (see #2662...

This seems even more important given that Xgboost plans to drop support from old binary format (see: [here](https://github.com/dmlc/xgboost/7547) and [here](https://github.com/dmlc/xgboost/blob/ee382c4153a675b0ecd21cc940e55ae8ae509a25/NEWS.md#new-serialization-format) ). It seems that SHAP used JSON [before](https://github.com/slundberg/shap/commit/397ed9278fc81c858d3188171dda7591531ccc0d) but it...

@rwjmiller did you find a possible quick solution to this problem? @dataversenomad I'm not sure that this is true using, `tree_method="approx"` or `"hist"` gives me the same error with `xgb.train`....

@githubnemo Yes this is exactly what I'm saying. Once in this new embedded space, one could use any algorithm in sklearn for classification / regression / clustering with much less...

@thomasjpfan that's exactly what I'm currently doing with a Semi Supervised VAE (M2), but as you say I think that there are many other applications. E.g. use a resnet and...

I don't think you can use training as a proxy (this is why I was talking about a `is_transform` flag). Indeed when training my transformer (imagine CNN or VAE) I...

@BenjaminBossan I agree with the downside from `is_transform`. I think what you propose makes sense 👍 : the only downside is that you have to pretrain outside of sklearn pipeline...

@benjamin-work that would be very useful for me also. I had some GAN type training to do, and always used some tricks to make it work in skorch. Essentially the...

Any news on this? That would be extremely helpful! Great library btw, thx :)

Ohh I somehow I missed that... Thanks, if you send a PR I'd be happy to merge it!