GradientBoosting with SKORCH

[francescamanni1989]

I would like to use skorch to implement Sklearn's GradientBoosting methods on torch based models.
Here is my first implementation on MNIST dataset, based on a previous solved issue on ADABoost for Skorch.

However I keep getting this error:

I tried to solve it, but I don't understand exactly why there is the incompatibility.

[thomasjpfan]

scikit-learn's GradientBoostingClassifier does not support changing the base estimator. The implementation only works with trees.

[BenjaminBossan]

Did you mean to use AdaBoostingClassifier?

[francescamanni1989]

Thanks for your prompt reply. I mean GradientBoostingClassifier. I though it could support a change on the base estimator by using Skorch since it does when using Ensemble-Pytorch.

[BenjaminBossan]

No, this does not work unfortunately. You can only change the loss, which should be a string, which is why you saw the error message above.

Given that we can't do anything about this situation on the skorch side, I'll close the issue. If you have further questions, feel free to re-open.

[francescamanni1989]

No, this does not work unfortunately. You can only change the loss, which should be a string, which is why you saw the error message above.

Given that we can't do anything about this situation on the skorch side, I'll close the issue. If you have further questions, feel free to re-open.

Is that the same with XGB classifier? Is it possible to adapt the base estimator and interface it with Skorch?