I'm using xgboost within H2O for a binary classification task. The dataset has several categorical features, to which the model applies a one-hot encoding during training.
Now I want to use SHAP (https://github.com/slundberg/shap) to locally interpret the predictions. For this, it would be nice to have the dataframe with the one-hot encoded columns and values. However, I seem to find no way to get this from the H2O model.
I could probably manually recreate the one-hot encoding, but maybe someone know a quicker solution?
