Which format does exported model use?

[ChangweiZhang]

can I get a onnx model by using ML.NET?

[GalOshri]

Not yet, but exporting ML.NET models to ONNX is on the roadmap. Is there a particular model/pipeline that you need to export to ONNX?

[ChangweiZhang]

According to the API reference, the class PredictionModel have a WriteAsync method and I can write model to file in binary format. What is the format?

[justinormont]

The format is a binary descriptor of the transforms and learner for the trained pipeline; create one and unzip it. As @GalOshri says, the ONNX output format is on the roadmap. We're always interested in what folks are looking for; any particular interest in ONNX? Or mainly just want to explore the output formats?

[GalOshri]

Adding to what @justinormont mentioned: the WriteAsync method saves the model as a zip file. You can then use ReadAsync to use the model in a different project.

[turowicz]

Interoperability with models trained in on different stack?

[GalOshri]

@turowicz which stack do you have in mind?

[enghch]

I'd like onnx to make it easy to consume my model with Windows ML to get the automatic hardware acceleration detection.