ONNX benchmarks

The following benchmarks compare runtime or backend with ONNX.

Benchmark for ONNX

The following benchmarks measure the prediction time between scikit-learn, onnxruntime and mlprodict for different models related to one-off predictions and batch predictions.

• Benchmark (ONNX) for common datasets (classification)
• Benchmark (ONNX) for common datasets (regression)
• Benchmark (ONNX) for common datasets (regression) with k-NN
• Benchmark (ONNX) for ensemble models
• Benchmark (ONNX) for LogisticRegression
• Benchmark (ONNX) for DecisionTreeClassifier
• Benchmark (ONNX) for DecisionTreeRegressor
• Benchmark (ONNX) for GradientBoostingRegressor
• Benchmark (ONNX) for GaussianProcessRegressor
• Benchmark (ONNX) for HistBoostingGradientRegressor
• Benchmark (ONNX) for KNeighborsClassifier
• Benchmark (ONNX) for RandomForestClassifier
• Benchmark (ONNX) for MLPClassifier

Benchmark specific operators (Add, Scaler, …)

The following benchmarks look into simplified models to help understand how runtime behave for specific operators.

• Benchmark (ONNX) for sklearn-onnx unit tests
• Benchmark (ONNX) for Addition
• Benchmark (ONNX) for Scaler
• Benchmark (ONNX) for intermediate outputs of a MLPClassifier

Profiling ONNX runtime

• Profiling predictions functions
  • Overview
  • Results

Full benchmarks with asv

Others benchmarks with asv and onnx:

• Scikit-Learn/ONNX benchmark with AirSpeedVelocity (official)

• Prediction with scikit-learn and ONNX benchmark

• Prediction with scikit-learn and ONNX benchmark (SVM + Trees)