How to export RegressionEnsemble to ONNX.

exportONNXNetwork only works on neural nets. However ONNX has support for regression models as demonstrated here: https://onnx.ai/sklearn-onnx/auto_examples/plot_convert_model.html

Anyone have any ideas on a workflow to get our model(s) out of matlab for use in Triton? There will probably need to be an intermediary format.

Answer 1

[Garmit Pant]

Hello Michael

From what I understand, you want to export MATLAB ‘RegressionEnsemble’ model to use with Triton.

Your intuition to use an intermediary format is correct. However, MATLAB currently doesn’t support the conversion of ensemble regression learner models like ‘RegressionEnsemble’ to any other format.

“exportONNXNetwork” function expects a ‘dlnetwork’ object as the input. Converting a ‘RegressionEnsemble’ model to a ‘dlnetwork’ in MATLAB is not possible because they represent fundamentally different types of models. ‘RegressionEnsemble’ is an ensemble of decision trees, whereas ‘dlnetwork’ is used for deep learning models.

If you need to deploy only ensemble learning models in Triton, then using scikit-learn in Python will be more suitable for your use case.

Scikit-learn in Python supports exporting multiple models, including ensemble learning models, to ONNX. The following resource provides a list of models supported by “skl2onnx”.

https://onnx.ai/sklearn-onnx/supported.html

You can also consider using a neural network for your task and exporting that from MATLAB using “exportONNXNetwork”.

You can refer the following MATLAB documentation for further understanding how to use the ‘dlnetwork’ object to design deep learning network for various tasks and workflows:

https://www.mathworks.com/help/releases/R2023b/deeplearning/ref/dlnetwork.html

I hope you find the above explanation and suggestions useful!