Source code for mlprodict.tools.asv_options_helper
"""
Functions to show shortened options in :epkg:`asv` benchmarks.
:githublink:`%|py|5`
"""
[docs]def expand_onnx_options(model, optim):
"""
Expands shortened options. Long names hide some part
of graphs in :epkg:`asv` benchmark. This trick converts
a string into real conversions options.
:param model: model class (:epkg:`scikit-learn`)
:param optim: option
:return: expanded options
It is the reverse of function :func:`shorten_onnx_options <mlprodict.tools.asv_options_helper.shorten_onnx_options>`.
The following options are handled:
.. runpython::
:showcode:
from sklearn.linear_model import LogisticRegression
from mlprodict.tools.asv_options_helper import expand_onnx_options
for name in ['cdist', 'nozipmap', 'raw_scores']:
print(name, ':', expand_onnx_options(LogisticRegression, name))
:githublink:`%|py|28`
"""
if optim == 'cdist':
options = {model.__class__: {'optim': 'cdist'}}
elif optim == 'nozipmap':
options = {model.__class__: {'zipmap': False}}
elif optim == 'raw_scores':
options = {model.__class__: {'raw_scores': True, 'zipmap': False}}
else:
options = optim # pragma: no cover
return options
[docs]def shorten_onnx_options(model, opts):
"""
Shortens onnx options into a string.
Long names hide some part
of graphs in :epkg:`asv` benchmark.
:param model: model class (:epkg:`scikit-learn`)
:param opts: options
:return: shortened options
It is the reverse of function :func:`expand_onnx_options <mlprodict.tools.asv_options_helper.expand_onnx_options>`.
:githublink:`%|py|51`
"""
if opts is None:
return opts
if opts == {model: {'optim': 'cdist'}}:
return 'cdist'
if opts == {model: {'zipmap': False}}:
return 'nozipmap'
if opts == {model: {'raw_scores': True, 'zipmap': False}}:
return 'raw_scores'
return None
[docs]def benchmark_version():
"""
Returns the list of ONNX version to benchmarks.
Following snippet of code shows which version is
current done.
.. runpython::
:showcode:
from mlprodict.tools.asv_options_helper import benchmark_version
print(benchmark_version())
:githublink:`%|py|74`
"""
return [13]
[docs]def ir_version():
"""
Returns the preferred `IR_VERSION
<https://github.com/onnx/onnx/blob/master/docs/IR.md#onnx-versioning>`_.
.. runpython::
:showcode:
from mlprodict.tools.asv_options_helper import ir_version
print(ir_version())
:githublink:`%|py|88`
"""
return [7]
[docs]def get_opset_number_from_onnx(benchmark=True):
"""
Retuns the current :epkg:`onnx` opset
based on the installed version of :epkg:`onnx`.
:param benchmark: returns the latest
version usable for benchmark
@eturn opset number
:githublink:`%|py|100`
"""
if benchmark:
return benchmark_version()[-1]
from onnx.defs import onnx_opset_version # pylint: disable=W0611
return onnx_opset_version()
[docs]def get_ir_version_from_onnx(benchmark=True):
"""
Retuns the current :epkg:`onnx` :epkg:`IR_VERSION`
based on the installed version of :epkg:`onnx`.
:param benchmark: returns the latest
version usable for benchmark
@eturn opset number
.. faqref::
:title: Failed to load model with error: Unknown model file format version.
:lid: l-onnx-ir-version-fail
:epkg:`onnxruntime` (or ``runtime='onnxruntime1'`` with :class:`OnnxInference <mlprodict.onnxrt.onnx_inference.OnnxInference>`)
fails sometimes to load a model showing the following error messsage:
::
RuntimeError: Unable to create InferenceSession due to '[ONNXRuntimeError] :
2 : INVALID_ARGUMENT : Failed to load model with error: Unknown model file format version.'
This case is due to metadata ``ir_version`` which defines the
:epkg:`IR_VERSION` or *ONNX version*. When a model is machine learned
model is converted, it is usually done with the default version
(``ir_version``) returned by the :epkg:`onnx` package.
:epkg:`onnxruntime` raises the above mentioned error message
when this version (``ir_version``) is too recent. In this case,
:epkg:`onnxruntime` should be updated to the latest version
available or the metadata ``ir_version`` can just be changed to
a lower number. Th function :func:`get_ir_version_from_onnx <mlprodict.tools.asv_options_helper.get_ir_version_from_onnx>`
returns the latest tested version with *mlprodict*.
.. runpython::
:showcode:
from sklearn.linear_model import LinearRegression
from sklearn.datasets import load_iris
from mlprodict.onnxrt import OnnxInference
import numpy
iris = load_iris()
X = iris.data[:, :2]
y = iris.target
lr = LinearRegression()
lr.fit(X, y)
# Conversion into ONNX.
from mlprodict.onnx_conv import to_onnx
model_onnx = to_onnx(lr, X.astype(numpy.float32),
target_opset=12)
print("ir_version", model_onnx.ir_version)
# Change ir_version
model_onnx.ir_version = 6
# Predictions with onnxruntime
oinf = OnnxInference(model_onnx, runtime='onnxruntime1')
ypred = oinf.run({'X': X[:5].astype(numpy.float32)})
print("ONNX output:", ypred)
# To avoid keep a fixed version number, you can use
# the value returned by function get_ir_version_from_onnx
from mlprodict.tools import get_ir_version_from_onnx
model_onnx.ir_version = get_ir_version_from_onnx()
print("ir_version", model_onnx.ir_version)
:githublink:`%|py|172`
"""
if benchmark:
return ir_version()[-1]
from onnx import IR_VERSION # pylint: disable=W0611
return IR_VERSION
[docs]def display_onnx(model_onnx, max_length=1000):
"""
Returns a shortened string of the model.
:param model_onnx: onnx model
:param max_length: maximal string length
:return: string
:githublink:`%|py|186`
"""
res = str(model_onnx)
if max_length is None or len(res) <= max_length:
return res
begin = res[:max_length // 2]
end = res[-max_length // 2:]
return "\n".join([begin, '[...]', end])
[docs]def version2number(vers):
"""
Converts a version number into a real number.
:githublink:`%|py|198`
"""
spl = vers.split('.')
r = 0
for i, s in enumerate(spl):
try:
vi = int(s)
except ValueError:
vi = 0
r += vi * 10 ** (-i * 3)
return r
