Xavier Dupré

+ @BowenBao who maintains the library as well. I see two ways of doing it. I think we need to add an extra parameter to let the user specify which...

I tried to replicate your issue with the following example and the conversion succeeds. Between step 2 and step 4, the model seems to be trained twice, the first instance...

I'm not sure I follow you what you are doing. RepeatedKFold produces several models, all different. All of them will be different from the model you convert in Step 3....

That's where I'm confused. The model is trained again in Step 3 and that would yield a different model and different prediction.

You can follow the examples from the tutorial : http://onnx.ai/sklearn-onnx/tutorial_2_new_converter.html. The most difficult part is to convert the transformation function [_transform_col](https://github.com/scikit-learn/scikit-learn/blob/main/sklearn/preprocessing/_data.py#L2635) into ONNX.

This PR may help you: https://github.com/onnx/sklearn-onnx/pull/404. It shows places in the code which need to be modified. A converter and a shape calculator must be implemented and registered. You may...

We first need onnx to support python 3.10. That should be the case for the next release. Then onnxruntime and finally sklearn-onnx.

I'm closing this issue. onnx and onnxruntime are now released on python 3.10.

int8 is not very common in standard machine learning, there is nothing preventing us from doing that except the converting library does not have a type Int8TensorType. It needs to...

Support for Int8 was added in the latest release. Could you try again?