"""
Rewrites some of the converters implemented in
:epkg:`sklearn-onnx`.
:githublink:`%|py|6`
"""
import numpy
from skl2onnx.operator_converters.decision_tree import (
convert_sklearn_decision_tree_regressor,
convert_sklearn_decision_tree_classifier)
from skl2onnx.operator_converters.gradient_boosting import (
convert_sklearn_gradient_boosting_regressor,
convert_sklearn_gradient_boosting_classifier)
from skl2onnx.operator_converters.random_forest import (
convert_sklearn_random_forest_classifier,
convert_sklearn_random_forest_regressor_converter)
from skl2onnx.common.data_types import guess_numpy_type
[docs]def _op_type_domain_regressor(dtype):
"""
Defines *op_type* and *op_domain* based on `dtype`.
:githublink:`%|py|22`
"""
if dtype == numpy.float32:
return 'TreeEnsembleRegressor', 'ai.onnx.ml', 1
if dtype == numpy.float64:
return 'TreeEnsembleRegressorDouble', 'mlprodict', 1
raise RuntimeError( # pragma: no cover
"Unsupported dtype {}.".format(dtype))
[docs]def _op_type_domain_classifier(dtype):
"""
Defines *op_type* and *op_domain* based on `dtype`.
:githublink:`%|py|34`
"""
if dtype == numpy.float32:
return 'TreeEnsembleClassifier', 'ai.onnx.ml', 1
if dtype == numpy.float64:
return 'TreeEnsembleClassifierDouble', 'mlprodict', 1
raise RuntimeError( # pragma: no cover
"Unsupported dtype {}.".format(dtype))
[docs]def new_convert_sklearn_decision_tree_classifier(scope, operator, container):
"""
Rewrites the converters implemented in
:epkg:`sklearn-onnx` to support an operator supporting
doubles.
:githublink:`%|py|48`
"""
dtype = guess_numpy_type(operator.inputs[0].type)
if dtype != numpy.float64:
dtype = numpy.float32
op_type, op_domain, op_version = _op_type_domain_classifier(dtype)
convert_sklearn_decision_tree_classifier(
scope, operator, container, op_type=op_type, op_domain=op_domain,
op_version=op_version)
[docs]def new_convert_sklearn_decision_tree_regressor(scope, operator, container):
"""
Rewrites the converters implemented in
:epkg:`sklearn-onnx` to support an operator supporting
doubles.
:githublink:`%|py|63`
"""
dtype = guess_numpy_type(operator.inputs[0].type)
if dtype != numpy.float64:
dtype = numpy.float32
op_type, op_domain, op_version = _op_type_domain_regressor(dtype)
convert_sklearn_decision_tree_regressor(
scope, operator, container, op_type=op_type, op_domain=op_domain,
op_version=op_version)
[docs]def new_convert_sklearn_gradient_boosting_classifier(scope, operator, container):
"""
Rewrites the converters implemented in
:epkg:`sklearn-onnx` to support an operator supporting
doubles.
:githublink:`%|py|78`
"""
dtype = guess_numpy_type(operator.inputs[0].type)
if dtype != numpy.float64:
dtype = numpy.float32
op_type, op_domain, op_version = _op_type_domain_classifier(dtype)
convert_sklearn_gradient_boosting_classifier(
scope, operator, container, op_type=op_type, op_domain=op_domain,
op_version=op_version)
[docs]def new_convert_sklearn_gradient_boosting_regressor(scope, operator, container):
"""
Rewrites the converters implemented in
:epkg:`sklearn-onnx` to support an operator supporting
doubles.
:githublink:`%|py|93`
"""
dtype = guess_numpy_type(operator.inputs[0].type)
if dtype != numpy.float64:
dtype = numpy.float32
op_type, op_domain, op_version = _op_type_domain_regressor(dtype)
convert_sklearn_gradient_boosting_regressor(
scope, operator, container, op_type=op_type, op_domain=op_domain,
op_version=op_version)
[docs]def new_convert_sklearn_random_forest_classifier(scope, operator, container):
"""
Rewrites the converters implemented in
:epkg:`sklearn-onnx` to support an operator supporting
doubles.
:githublink:`%|py|108`
"""
dtype = guess_numpy_type(operator.inputs[0].type)
if dtype != numpy.float64:
dtype = numpy.float32
op_type, op_domain, op_version = _op_type_domain_classifier(dtype)
convert_sklearn_random_forest_classifier(
scope, operator, container, op_type=op_type, op_domain=op_domain,
op_version=op_version)
[docs]def new_convert_sklearn_random_forest_regressor(scope, operator, container):
"""
Rewrites the converters implemented in
:epkg:`sklearn-onnx` to support an operator supporting
doubles.
:githublink:`%|py|123`
"""
dtype = guess_numpy_type(operator.inputs[0].type)
if dtype != numpy.float64:
dtype = numpy.float32
op_type, op_domain, op_version = _op_type_domain_regressor(dtype)
convert_sklearn_random_forest_regressor_converter(
scope, operator, container, op_type=op_type, op_domain=op_domain,
op_version=op_version)
