OnnxConverter Class

The converter that converts the pipeline/model objects from pkl format to onnx format.

Before the input data is fitted/transformed, call Initialize to setup the signature of the X. Then the convert method can be called multi times, to convert given sklearn pipeline.

Construct the Onnx converter.

Inheritance

builtins.object

OnnxConverter

Constructor

OnnxConverter(version='', is_onnx_compatible=False, enable_split_onnx_featurizer_estimator_models=False, **kwargs)

Parameters

Name	Description
version
is_onnx_compatible	default value: False
enable_split_onnx_featurizer_estimator_models	default value: False

Methods

convert	Convert the Python model with given child run object into ONNX model.
get_converted_onnx_model_resource	Get the resource of the converted ONNX model.
get_init_metadata_dict	Return the metadata dictionary of the onnx converter in the init phase.
get_onnx_metadata	given the training data, generate metadata needed for ONNX conversion
initialize_input	Initialize the converter with given training data X. NOTE.The X must be the original value, before it's transformed. :param X: The original input X before transformation applied.
initialize_with_metadata	Initialize the onnx converter with the previously generated init phase metadata.
is_initialized	Return if the converter is initialized or not.
load_onnx_model	Load the binary format ONNX model.
save_onnx_model	Save an ONNX model to a ProtoBuf object binary file.
save_onnx_model_to_text_file	Save the ONNX model in text format.

convert

Convert the Python model with given child run object into ONNX model.

convert(raw_model: Pipeline, model_name: str = '', model_desc: str | Dict[str, Any] | None = None) -> Tuple[ModelProto | None, ModelProto | None, ModelProto | None, Dict[str, Any] | None]

Parameters

Name	Description
raw_model Required	The sklearn pipeline object to convert to ONNX.
model_name Required	The name of the model.
model_desc	The model description. default value: None

get_converted_onnx_model_resource

Get the resource of the converted ONNX model.

get_converted_onnx_model_resource() -> Dict[Any, Any]

get_init_metadata_dict

Return the metadata dictionary of the onnx converter in the init phase.

get_init_metadata_dict() -> Dict[str, Any] | None

get_onnx_metadata

given the training data, generate metadata needed for ONNX conversion

static get_onnx_metadata(X: ndarray | DataFrame | spmatrix | EnginelessDataflow, x_raw_column_names: ndarray | None = None) -> Dict[str, Any] | None

Parameters

Name	Description
X Required	Input data.
x_raw_column_names	default value: None

Returns

Type	Description
	A metadata dictionary used by ONNX conversion.

initialize_input

Initialize the converter with given training data X.

NOTE.The X must be the original value, before it's transformed. :param X: The original input X before transformation applied.

initialize_input(X: ndarray | DataFrame, x_raw_column_names: ndarray | None = None, model_name: str = '', model_desc: Dict[Any, Any] | None = None) -> None

Parameters

Name	Description
X Required
x_raw_column_names	default value: None
model_name
model_desc	default value: None

initialize_with_metadata

Initialize the onnx converter with the previously generated init phase metadata.

initialize_with_metadata(metadata_dict: Dict[str, Any] | None, model_name: str = '', model_desc: Dict[Any, Any] | None = None) -> None

Parameters

Name	Description
metadata_dict Required
model_name
model_desc	default value: None

is_initialized

Return if the converter is initialized or not.

is_initialized()

load_onnx_model

Load the binary format ONNX model.

static load_onnx_model(file_path: str) -> ModelProto

Parameters

Name	Description
file_path Required	the path including file name to save the model

Returns

Type	Description
	the loaded onnx model object.

save_onnx_model

Save an ONNX model to a ProtoBuf object binary file.

static save_onnx_model(onnx_model: ModelProto, file_path: str) -> None

Parameters

Name	Description
model Required	ONNX model
file_path Required	ONNX file (full file name)
onnx_model Required

save_onnx_model_to_text_file

Save the ONNX model in text format.

static save_onnx_model_to_text_file(onnx_model: ModelProto, file_path: str) -> None

Parameters

Name	Description
model Required	ONNX model (object)
file_path Required	the path including file name to save the model
onnx_model Required