Implementation:Onnx Onnx ML Operator Sets

Knowledge Sources	Onnx_Onnx
Domains	Machine Learning Operators, Schema Registration
Last Updated	2026-02-10 00:00 GMT

Overview

Concrete tool for declaring and registering ONNX ML domain (ai.onnx.ml) operator schemas across versions 1 through 5, provided by the ONNX library.

Description

This header defines the operator schema forward declarations and versioned operator set classes for the ONNX ML domain. It covers traditional machine learning operators that extend beyond standard tensor operations, targeting classical ML model types such as classifiers, regressors, and feature transformers.

The file is organized by operator set version:

Version 1 declares 18 operators: ArrayFeatureExtractor, Binarizer, CastMap, CategoryMapper, DictVectorizer, FeatureVectorizer, Imputer, LabelEncoder, LinearClassifier, LinearRegressor, Normalizer, OneHotEncoder, SVMClassifier, SVMRegressor, Scaler, TreeEnsembleClassifier, TreeEnsembleRegressor, and ZipMap.

Version 2 introduces an updated LabelEncoder.

Version 3 updates TreeEnsembleClassifier and TreeEnsembleRegressor.

Version 4 introduces another updated LabelEncoder.

Version 5 introduces the new TreeEnsemble operator alongside updated TreeEnsembleRegressor and TreeEnsembleClassifier.

Each version is encapsulated in an OpSet_OnnxML_verN class with a static ForEachSchema method that iterates over the schemas for that version. The RegisterOnnxMLOperatorSetSchema function registers all versions into the global schema registry. The entire file is guarded by the ONNX_ML preprocessor macro, so these operators are only compiled when ML support is enabled.

Usage

Use this header when building ONNX with ML operator support enabled (ONNX_ML defined). The RegisterOnnxMLOperatorSetSchema function should be called during library initialization to make all ML operators available for model validation, shape inference, and other schema-dependent operations. This is essential for processing ONNX models exported from scikit-learn, XGBoost, LightGBM, and similar ML frameworks via converters like sklearn-onnx.

Code Reference

Source Location

• Repository: Onnx_Onnx
• File: onnx/defs/operator_sets_ml.h

Signature

class OpSet_OnnxML_ver1 {
public:
    static void ForEachSchema(const std::function<void(OpSchema&&)>& fn);
};

class OpSet_OnnxML_ver2 {
public:
    static void ForEachSchema(const std::function<void(OpSchema&&)>& fn);
};

class OpSet_OnnxML_ver3 {
public:
    static void ForEachSchema(const std::function<void(OpSchema&&)>& fn);
};

class OpSet_OnnxML_ver4 {
public:
    static void ForEachSchema(const std::function<void(OpSchema&&)>& fn);
};

class OpSet_OnnxML_ver5 {
public:
    static void ForEachSchema(const std::function<void(OpSchema&&)>& fn);
};

inline void RegisterOnnxMLOperatorSetSchema();

Import

#include "onnx/defs/operator_sets_ml.h"

I/O Contract

Inputs (RegisterOnnxMLOperatorSetSchema)

Name	Type	Required	Description
(none)	-	-	This function takes no parameters; it registers all ML operator schemas into the global OpSchemaRegistry

Outputs

Name	Type	Description
(side effect)	void	All ML operator schemas from versions 1 through 5 are registered in the global schema registry

ML Operators by Version

Version 1

• ArrayFeatureExtractor, Binarizer, CastMap, CategoryMapper
• DictVectorizer, FeatureVectorizer, Imputer, LabelEncoder
• LinearClassifier, LinearRegressor, Normalizer, OneHotEncoder
• SVMClassifier, SVMRegressor, Scaler
• TreeEnsembleClassifier, TreeEnsembleRegressor, ZipMap

Version 2

• LabelEncoder (updated)

Version 3

• TreeEnsembleClassifier (updated), TreeEnsembleRegressor (updated)

Version 4

• LabelEncoder (updated)

Version 5

• TreeEnsemble (new), TreeEnsembleRegressor (updated), TreeEnsembleClassifier (updated)

Usage Examples

// Register all ML operator schemas at library initialization
#ifdef ONNX_ML
RegisterOnnxMLOperatorSetSchema();
#endif

// Iterate over version 1 ML operators
OpSet_OnnxML_ver1::ForEachSchema([](OpSchema&& schema) {
    // Process each ML operator schema
    std::cout << "ML Operator: " << schema.Name() << std::endl;
});

Related Pages

• Environment:Onnx_Onnx_Cpp_Build_Environment