Source code for mlprodict.onnxrt.ops_cpu.op_linear_regressor
# -*- encoding: utf-8 -*-
# pylint: disable=E0203,E1101,C0111
"""
Runtime operator.
:githublink:`%|py|7`
"""
import numpy
from ._op import OpRunUnaryNum
from ._op_numpy_helper import numpy_dot_inplace
[docs]class LinearRegressor(OpRunUnaryNum):
atts = {'coefficients': None, 'intercepts': None,
'targets': 1, 'post_transform': b'NONE'}
[docs] def __init__(self, onnx_node, desc=None, **options):
OpRunUnaryNum.__init__(self, onnx_node, desc=desc,
expected_attributes=LinearRegressor.atts,
**options)
if not isinstance(self.coefficients, numpy.ndarray):
raise TypeError( # pragma: no cover
"coefficient must be an array not {}.".format(
type(self.coefficients)))
n = self.coefficients.shape[0] // self.targets
self.coefficients = self.coefficients.reshape(self.targets, n).T
[docs] def _run(self, x): # pylint: disable=W0221
score = numpy_dot_inplace(self.inplaces, x, self.coefficients)
if self.intercepts is not None:
score += self.intercepts
if self.post_transform == b'NONE':
pass
else:
raise NotImplementedError( # pragma: no cover
"Unknown post_transform: '{}'.".format(
self.post_transform))
return (score, )