def test_guess_type(self):
dtypes = [
(np.int32, Int32TensorType),
(np.int64, Int64TensorType),
(np.float32, FloatTensorType),
(np.str_, StringTensorType),
(np.bool_, BooleanTensorType),
(np.int8, Int8TensorType),
(np.uint8, UInt8TensorType)
]
if Complex64TensorType is not None:
dtypes.append((np.complex64, Complex64TensorType))
if Complex128TensorType is not None:
dtypes.append((np.complex128, Complex128TensorType))
for dtype, exp in dtypes:
if dtype == np.str_:
mat = np.empty((3, 3), dtype=dtype)
mat[:, :] = ""
else:
mat = np.zeros((3, 3), dtype=dtype)
res = _guess_type(mat)
assert isinstance(res, exp)
dtypes = [np.float64]
for dtype in dtypes:
mat = np.zeros((3, 3), dtype=dtype)
_guess_type(mat, )
def guess_initial_types(X, initial_types):
"""
Guesses initial types from an array or a dataframe.
@param X array or dataframe
@param initial_types hints about X
@return data types
"""
if X is None and initial_types is None:
raise NotImplementedError( # pragma: no cover
"Initial types must be specified.")
elif initial_types is None:
if isinstance(X, (numpy.ndarray, pandas.DataFrame)):
X = X[:1]
if isinstance(X, pandas.DataFrame):
initial_types = []
for c in X.columns:
if isinstance(X[c].values[0], (str, numpy.str_)):
g = StringTensorType()
else:
g = _guess_type(X[c].values)
g.shape = [None, 1]
initial_types.append((c, g))
else:
gt = _guess_type(X)
initial_types = [('X', gt)]
return initial_types
def _display_intermediate_steps(model_onnx, inputs):
import onnxruntime
print("[_display_intermediate_steps] BEGIN")
if isinstance(model_onnx, str):
import onnx
model_onnx = onnx.load(model_onnx)
for name, node in enumerate_model_initializers(model_onnx, add_node=True):
print("INIT: {} - {}".format(name, _guess_type(node)))
for out, node in enumerate_model_node_outputs(model_onnx, add_node=True):
print('-')
print("OUTPUT: {} from {}".format(out, node.name))
step = select_model_inputs_outputs(model_onnx, out)
try:
step_sess = onnxruntime.InferenceSession(step.SerializeToString())
except Exception as e:
raise RuntimeError("Unable to load ONNX model with onnxruntime. "
"Last added node is:\n{}".format(node)) from e
for o in step_sess.get_inputs():
print("IN :", o)
for o in step_sess.get_outputs():
print("OUT: ", o)
if inputs:
res = step_sess.run(inputs)
print(res)
print("[_display_intermediate_steps] END")
def _display_intermediate_steps(model_onnx, inputs, disable_optimisation):
import onnxruntime
print("[_display_intermediate_steps] BEGIN")
if isinstance(model_onnx, str):
import onnx
model_onnx = onnx.load(model_onnx)
for name, node in enumerate_model_initializers(model_onnx, add_node=True):
print("INIT: {} - {}".format(name, _guess_type(node)))
for out, node in enumerate_model_node_outputs(model_onnx, add_node=True):
print('-')
print("OUTPUT: {} from {}".format(out, node.name))
step = select_model_inputs_outputs(model_onnx, out)
if (disable_optimisation
and hasattr(onnxruntime, 'GraphOptimizationLevel')):
opts = onnxruntime.SessionOptions()
opts.graph_optimization_level = (
onnxruntime.GraphOptimizationLevel.ORT_DISABLE_ALL)
else:
opts = None
try:
step_sess = onnxruntime.InferenceSession(step.SerializeToString(),
sess_options=opts)
except Exception as e:
raise RuntimeError("Unable to load ONNX model with onnxruntime. "
"Last added node is:\n{}".format(node)) from e
for o in step_sess.get_inputs():
print("IN :", o)
for o in step_sess.get_outputs():
print("OUT: ", o)
if inputs:
res = step_sess.run(inputs)
print(res)
print("[_display_intermediate_steps] END")
def test_guess_type(self):
dtypes = [(np.int32, Int32TensorType), (np.int64, Int64TensorType),
(np.float32, FloatTensorType), (np.str, StringTensorType)]
for dtype, exp in dtypes:
if dtype == np.str:
mat = np.empty((3, 3), dtype=dtype)
mat[:, :] = ""
else:
mat = np.zeros((3, 3), dtype=dtype)
res = _guess_type(mat)
assert isinstance(res, exp)
dtypes = [np.float64]
for dtype in dtypes:
mat = np.zeros((3, 3), dtype=dtype)
_guess_type(mat, )
def test_guess_type(self):
dtypes = [(np.int32, Int32TensorType), (np.int64, Int64TensorType),
(np.float32, FloatTensorType), (np.str, StringTensorType)]
for dtype, exp in dtypes:
if dtype == np.str:
mat = np.empty((3, 3), dtype=dtype)
mat[:, :] = ""
else:
mat = np.zeros((3, 3), dtype=dtype)
res = _guess_type(mat)
assert isinstance(res, exp)
dtypes = [np.float64]
for dtype in dtypes:
mat = np.zeros((3, 3), dtype=dtype)
try:
_guess_type(mat, )
raise AssertionError("It should fail for type "
"{}".format(dtype))
except NotImplementedError:
pass