ONNX benchmarks

The following benchmarks compare runtime or backeend with ONNX.

One-Off predictions

The following benchmark measures the prediction time between scikit-learn and onnxruntime for different configurations related to one-off predictions: predictions are computed for one observation at a time which is the standard scenario in a webservice. onnxruntime allows for some models to run batch predictions. If this functionality is available, it is usually tested for small batches (like 10 observations).