Module onnx 

ONNX export and inference — train once; run in Rust, Python, or any ONNX runtime.

ExportOnnx writes a trained model — or a whole Pipeline — to a single .onnx file via onnx-export-rs. InferenceModel loads any ONNX file and runs it: linear / NN graphs through tract, and the ONNX-ML tree-ensemble ops tract doesn’t implement (from an exported forest) through a small native interpreter. So the exported artifact always round-trips back into Rust — a RandomForest included — and stays portable to every other ONNX runtime.

A pipeline is exported by splicing each leading transformer in front of the estimator’s graph, in order — scalers as an affine (x - shift) / scale, imputers as Where(IsNaN(x), fill, x), one-hot encoders as Concat(Cast(Equal(Round(Gather(x)), cat))) — so the result is one self-contained graph: raw features in, predictions out. A step with no ONNX form is reported as an error naming it.

Structs

InferenceModel

A loaded ONNX model, ready to run.

Enums

Prefix

A preprocessing step expressible as ONNX graph nodes, prepended in front of an estimator so a whole pipeline becomes one graph. Each maps a same-width feature tensor to another.

Traits

ExportOnnx

A model or pipeline that can be exported to ONNX.