Xgboost conversion to ONNX error

[chengnignzhang]

Hi

I am facing below error
ValueError: No proper operator name found for '<class 'xgboost.core.Booster'>'
while converting my xgboost model to ONNX. The model was trained using xgboost.

used convert_xgboost
onx = onnxmltools.convert.convert_xgboost(loaded_model, initial_types=initial_type)
Any thoughts on this error?

Does this mean xgboost.core.Booster is not supported for now?

Thanks
Chengning

[jiafatom]

cc @xadupre @prabhat00155

[xadupre]

It is not. This conversion is implemented for LightGbm and not for XGBoost. We could replicate the same trick if needed.

[ankit1200]

@xadupre do you have an example on how this would work? I'm trying to convert a xgb model to onnx and getting a similar error.

Error
ValueError: No proper operator name found for '<class 'xgboost.core.Booster'>'

Code

    initial_types = [('float_input', FloatTensorType([1, len(features)]))]
    onnx_model = convert_xgboost(xgb_model,'tree-based classifier', initial_types = initial_types)
    
    with open("model_files/test_xgb.onnx", "wb") as f:
        f.write(onnx_model.SerializeToString())

[xadupre]

I wrote a unit test to make sure it was passing: https://github.com/onnx/onnxmltools/blob/master/tests/xgboost/test_xgboost_converters.py#L188. How do you train your model?

[ankit1200]

@xadupre We're using dask_xgboost: <class 'dask_xgboost.core.XGBRegressor'>. I've tried to convert the dask_xgboost model to onnx and the booster too: <class 'xgboost.core.Booster'>. Your help on this issue will be much appreciated