Correct way of converting HistGradientBoostingRegressor with categorical_features?

Open ogencoglu opened this issue 1 year ago • 2 comments

HistGradientBoostingRegressor supports specification of categorical_features: https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.HistGradientBoostingRegressor.html#sklearn.ensemble.HistGradientBoostingRegressor

What would be the correct initial_type in convert_sklearn(model, initial_types) to achieve this behaviour. Float inputs to ONNX are giving wrong predictions of course.

Jun 04 '24 13:06 ogencoglu

Pinging @maximilianeber in case you found out a way

Jun 04 '24 13:06 ogencoglu

Pinging @maximilianeber in case you found out a way

Replied in the original thread in https://github.com/onnx/sklearn-onnx/issues/1051. We haven't managed to work around this and stuck to OHE for now.

Jun 04 '24 13:06 maximilianeber