"""
Inspired from skl2onnx, handles two backends.
:githublink:`%|py|5`
"""
from onnxruntime import (
InferenceSession, GraphOptimizationLevel, SessionOptions)
from pyquickhelper.pycode import is_travis_or_appveyor
from .utils_backend_common_compare import compare_runtime_session
[docs]def _capture_output(fct, kind):
if is_travis_or_appveyor():
return fct(), None, None # pragma: no cover
try:
from cpyquickhelper.io import capture_output
except ImportError:
# cpyquickhelper not available
return fct(), None, None # pragma: no cover
return capture_output(fct, kind) # pragma: no cover
[docs]class InferenceSession2:
"""
Overwrites class *InferenceSession* to capture
the standard output and error.
:githublink:`%|py|26`
"""
[docs] def __init__(self, *args, **kwargs):
"Overwrites the constructor."
runtime_options = kwargs.pop('runtime_options', {})
disable_optimisation = runtime_options.pop(
'disable_optimisation', False)
if disable_optimisation:
if 'sess_options' in kwargs:
raise RuntimeError( # pragma: no cover
"Incompatible options, 'disable_options' and 'sess_options' cannot "
"be sepcified at the same time.")
kwargs['sess_options'] = SessionOptions()
kwargs['sess_options'].graph_optimization_level = (
GraphOptimizationLevel.ORT_DISABLE_ALL)
self.sess, self.outi, self.erri = _capture_output(
lambda: InferenceSession(*args, **kwargs), 'c')
[docs] def run(self, *args, **kwargs):
"Overwrites method *run*."
res, self.outr, self.errr = _capture_output(
lambda: self.sess.run(*args, **kwargs), 'c')
return res
[docs] def get_outputs(self, *args, **kwargs):
"Overwrites method *get_outputs*."
return self.sess.get_outputs(*args, **kwargs)
[docs]def compare_runtime(test, decimal=5, options=None,
verbose=False, context=None, comparable_outputs=None,
intermediate_steps=False, classes=None,
disable_optimisation=False):
"""
The function compares the expected output (computed with
the model before being converted to ONNX) and the ONNX output
produced with module :epkg:`onnxruntime` or :epkg:`mlprodict`.
:param test: dictionary with the following keys:
- *onnx*: onnx model (filename or object)
- *expected*: expected output (filename pkl or object)
- *data*: input data (filename pkl or object)
:param decimal: precision of the comparison
:param options: comparison options
:param context: specifies custom operators
:param verbose: in case of error, the function may print
more information on the standard output
:param comparable_outputs: compare only these outputs
:param intermediate_steps: displays intermediate steps
in case of an error
:param classes: classes names (if option 'nocl' is used)
:param disable_optimisation: disable optimisation onnxruntime
could do
:return: tuple (outut, lambda function to run the predictions)
The function does not return anything but raises an error
if the comparison failed.
:githublink:`%|py|87`
"""
return compare_runtime_session(
InferenceSession2, test, decimal=decimal, options=options,
verbose=verbose, context=context,
comparable_outputs=comparable_outputs,
intermediate_steps=intermediate_steps,
classes=classes, disable_optimisation=disable_optimisation)
