ONNX Test in Go

Steps to reproduce

Setup Python environment

conda create -n "onnx" python=3.12
conda activate onnx
conda install conda-forge::scikit-learn
conda install conda-forge::skl2onnx
conda install conda-forge::onnxruntime

Build models

python build/build.py

This will create:

• data/rf_iris.joblib the model persisted using Python joblib
• data/rf_iris.onnx the model persisted using ONNX skl2onnx
• data/X_test.csv the test data
• data/y_test.csv the test ground truth values

Test models

Setting up ONNX Runtime

You will need the ONNX runtime for your platform. Visit https://github.com/microsoft/onnxruntime/releases and download the release for your platform. Extract the downloaded files and copy the shared libary to the inference/go_onnx folder. On Intel Mac this was libonnxruntime.1.18.0.dylib. If necessary, update the following line in main.go:

ort.SetSharedLibraryPath("libonnxruntime.1.18.0.dylib")

Running the tests

Test using the model loaded from joblib:

python inference/python_sk/predict.py
Accuracy:  0.9736842105263158
Total run time: 150.890ms
Inference time: 4.618ms

Test using the model loaded from onnxruntime:

First run:

python inference/python_onnx/predict.py
Accuracy:  0.9736842105263158
Total run time: 83.337ms
Inference time: 3.119ms5

Subsequent runs:

python inference/python_onnx/predict.py
Accuracy:  0.9736842105263158
Total run time: 7.824ms
Inference time: 1.015ms

Test using the model loaded in Go:

./onnx_test
Accuracy:  0.9736842105263158
Total run time:  34.424473ms
Inference time:  395.164µs