def test_onnx_example_constant_of_shape(self):
x = np.array([1, 2, 4, 5, 5, 4]).astype(np.float32).reshape((3, 2))
opv = _TARGET_OPSET_
cop2 = OnnxConstantOfShape(OnnxShape('input', op_version=opv),
output_names=['mat'],
op_version=opv)
model_def = cop2.to_onnx({'input': x},
outputs=[('mat', FloatTensorType())])
sess = InferenceSession(model_def.SerializeToString())
res = sess.run(None, {'input': x})
exp = np.zeros((3, 2), dtype=np.float32)
assert_almost_equal(exp, res[0])
tensor_value = onnx.helper.make_tensor("value", onnx.TensorProto.FLOAT,
(1, ), [-5])
cop2 = OnnxConstantOfShape(OnnxShape('input', op_version=opv),
value=tensor_value,
output_names=['mat'],
op_version=opv)
model_def = cop2.to_onnx({'input': x},
outputs=[('mat', FloatTensorType())])
sess = InferenceSession(model_def.SerializeToString())
res = sess.run(None, {'input': x})
exp = np.full((3, 2), -5.)
assert_almost_equal(exp, res[0])
def test_onnxt_runtime_constant_of_shape(self):
x = numpy.array([2, 2], dtype=numpy.int64)
y = numpy.zeros((2, 2))
onx = OnnxConstantOfShape('X', output_names=['Y'])
model_def = onx.to_onnx({'X': x.astype(numpy.float32)},
outputs=[('Y', FloatTensorType())])
got = OnnxInference(model_def).run({'X': x})
self.assertEqualArray(y, got['Y'])
def test_constant_of_shape(self):
for opset in range(20, 8, -1):
if opset > TARGET_OPSET:
continue
for value in [
np.array([5], dtype=np.float32),
np.array(5, dtype=np.float32)
]:
with self.subTest(opset=opset, value=value):
tensor_value = onnx.helper.make_tensor(
"value", onnx.TensorProto.FLOAT, [1], [5])
cst = OnnxConstantOfShape('X',
value=tensor_value,
op_version=opset,
output_names=['Y'])
shape = np.array([3, 4], dtype=np.int64)
onx = cst.to_onnx({'X': shape},
target_opset=opset,
outputs=[('Y', FloatTensorType())])
sess = InferenceSession(onx.SerializeToString())
res = sess.run(None, {'X': shape})
assert_almost_equal(
res[0], np.full(tuple(shape), 5, dtype=np.float32))
cst = OnnxConstantOfShape('X',
value=value,
op_version=opset,
output_names=['Y'])
shape = np.array([3, 4], dtype=np.int64)
onx = cst.to_onnx({'X': shape},
target_opset=opset,
outputs=[('Y', FloatTensorType())])
sess = InferenceSession(onx.SerializeToString())
res = sess.run(None, {'X': shape})
assert_almost_equal(
res[0], np.full(tuple(shape), 5, dtype=np.float32))
for opset in [TARGET_OPSET]:
for value in [5, np.float32(5)]:
with self.subTest(opset=opset, value=value):
with self.assertRaises(TypeError):
OnnxConstantOfShape('X',
value=value,
op_version=opset,
output_names=['Y'])
def test_constant_of_shape(self):
x = numpy.array([1, 2, 4, 5, 5, 4]).astype(numpy.float32).reshape(
(3, 2))
tensor_value = make_tensor("value", TensorProto.FLOAT, (1, ), [-5]) # pylint: disable=E1101
cop2 = OnnxConstantOfShape(OnnxShape('input'),
value=tensor_value,
output_names=['mat'])
model_def = cop2.to_onnx({'input': x},
outputs=[('mat', FloatTensorType())])
oinf = OnnxInference(model_def, skip_run=True)
dot = oinf.to_dot()
self.assertIn('ConstantOfShape', dot)