Mustafa Bal

It seems like this issue stems from the ML.NET end, as an issue for this had already been created before. ([Issue #1109 on dotnet/machinelearning](https://github.com/dotnet/machinelearning/issues/1109)).

This is the code snipped that should be taking care of hidden fields that are coming from ML.NET to NimbusML: https://github.com/microsoft/NimbusML/blob/56bbda6cbe02c1fe79fa48d9e756be2450720a10/src/DotNetBridge/NativeDataInterop.cs#L127-L128

The exact problem that this issue addresses is as follows: Add support to DatasetTransformer so that it can **use model files** that contain predictors by calling the `.transform()` function. So...

Hi @can-anyscale @anyscalesam, any updates on this? The Ray project is very much being held back without Python 3.12 support.

@VivekPanyam I also agree with your assessment on using [pykeio/ort](https://github.com/pykeio/ort) by default and having `wonnx` as an explicit opt-in. There can be a case made for having `wonnx` as default...