Samples of ONNX export

How do I create ONNX files?

Below, several ways of exporting certain models as ONNX from different frameworks are shown using examples. The samples do not claim to be complete and only serve to provide a primary overview. For more detailed documentation, refer to the documentation for the respective framework.

The listed examples are limited to the creation of an ONNX file. Examples for conversion to make the file available in TwinCAT can be found here: Converting ONNX to XML and BML as well as in the ZIP archive for the linked samples (see below) in the PythonAPI_mllib folder.

Overview of available samples

Python package	Model type	Option	Comment	Sample link
PyTorch	MLP Regressor			GoToPage
Keras	MLP Regressor			GoToPage
Scikit-learn	MLP Regressor			GoToPage
Scikit-learn	MLP Classifier		ONNX graph must be adapted	GoToPage
Scikit-learn	SVR			GoToPage
Scikit-learn	SVC	decision_function_shape='ovo'		GoToPage
Scikit-learn	k-means		Meta Key must be entered in ONNX.	GoToPage
Scikit-learn	PCA			GoToPage
Scikit-learn	Decision Tree Classifier			GoToPage
Scikit-learn	Decision Tree Regressor			GoToPage
Scikit-learn	Extra Tree Classifier			GoToPage
Scikit-learn	Extra Tree Regressor			GoToPage
Scikit-learn	Extra Trees Classifier			GoToPage
Scikit-learn	Extra Trees Regressor			GoToPage
Scikit-learn	Random Forest Classifier			GoToPage
Scikit-learn	Random Forest Regressor			GoToPage
LightGBM	Random Forest Regressor		ONNX graph must be adapted	GoToPage
Scikit-learn	Gradient Boosting Classifier			GoToPage
Scikit-learn	Gradient Boosting Regressor			GoToPage
Scikit-learn	Hist Gradient Boosting Classifier			GoToPage
Scikit-learn	Hist Gradient Boosting Regressor			GoToPage
XGBoost	XGBClassifier	Not all configurations allow an ONNX export	Package version <= 1.5.2 or >= 1.7.4 required	GoToPage
XGBoost	XCBRegressor	Not all configurations allow an ONNX export	Package version <= 1.5.2 or >= 1.7.4 required	GoToPage
LightGBM	LGBMRegressor	Not all configurations allow an ONNX export		GoToPage
LightGBM	LGBMClassifier		ONNX graph must be adapted	GoToPage

All samples can be downloaded here as a ZIP archive: Beckhoff_ONNX_Samples.zip