Add Support for XGBoost

[b-yogesh]

Hi @Julia310, while we were at the Science-Hub course, a team of students wanted to train a model using XGBoost and were asking the recommended way to train a model for the xarray datasets and we pointed them here. But then we realized that we currently do not support the xgboost library yet.

It would be a great addition to ml4xcube package.

Suggested solutions:

• We can use the xgboost's sklearn compatible interface and then just update the model type to be both BaseEstimator and XGBModel or something like that.
• We can also create a new Trainer class for XGBoost for its own interface.

Let me know which approach would be best here and if you think it would enhance ml4xcube for xarray-based ML applications.

[Julia310]

Hi @b-yogesh,

I was actually planning to remove the trainer classes from ml4xcube, as well-established alternatives like PyTorch Lightning and TensorFlow already provide similar interfaces that are widely adopted and easy to use. The focus of ml4xcube would then shift entirely to simplifying data preparation for these frameworks.

As part of this change, the notebooks would also be updated accordingly. In this context, adding an XGBoost example makes sense and could be a valuable addition.

What do you think about this approach? I'd love to hear your thoughts!

[b-yogesh]

Hi @Julia310 , yes I think that is a good idea to avoid another layer of abstraction for the Trainer class and make the focus of this package to make processing easier for data cubes that can then be used for training with different ML frameworks. Thank you!