PRelu

PRelu - 16

Version

• name: PRelu (GitHub)

• domain: main

• since_version: 16

• function: True

• support_level: SupportType.COMMON

• shape inference: True

This version of the operator has been available since version 16.

Summary

PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise.

History - Version 16 adds bfloat16 to the types allowed. This operator supports unidirectional broadcasting (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX.

Inputs

• X (heterogeneous) - T: Input tensor

• slope (heterogeneous) - T: Slope tensor. The shape of slope can be smaller then first input X; if so, its shape must be unidirectional broadcastable to X

Outputs

• Y (heterogeneous) - T: Output tensor (same size as X)

Type Constraints

• T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.

Examples

default

node = onnx.helper.make_node(
    "PRelu",
    inputs=["x", "slope"],
    outputs=["y"],
)

x = np.random.randn(3, 4, 5).astype(np.float32)
slope = np.random.randn(3, 4, 5).astype(np.float32)
y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * slope

expect(node, inputs=[x, slope], outputs=[y], name="test_prelu_example")

_prelu_broadcast

node = onnx.helper.make_node(
    "PRelu",
    inputs=["x", "slope"],
    outputs=["y"],
)

x = np.random.randn(3, 4, 5).astype(np.float32)
slope = np.random.randn(5).astype(np.float32)
y = np.clip(x, 0, np.inf) + np.clip(x, -np.inf, 0) * slope

expect(node, inputs=[x, slope], outputs=[y], name="test_prelu_broadcast")

Differences

0	0	PRelu takes input data (Tensor) and slope tensor as input, and produces one	PRelu takes input data (Tensor) and slope tensor as input, and produces one
1	1	output data (Tensor) where the function f(x) = slope * x for x < 0,	output data (Tensor) where the function f(x) = slope * x for x < 0,
2	2	f(x) = x for x >= 0., is applied to the data tensor elementwise.	f(x) = x for x >= 0., is applied to the data tensor elementwise.
	3
	4		**History**
	5		- Version 16 adds bfloat16 to the types allowed.
3	6	This operator supports **unidirectional broadcasting** (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX _.	This operator supports **unidirectional broadcasting** (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX _.
4	7
5	8	**Inputs**	**Inputs**
6	9
7	10	* **X** (heterogeneous) - **T**:	* **X** (heterogeneous) - **T**:
8	11	Input tensor	Input tensor
9	12	* **slope** (heterogeneous) - **T**:	* **slope** (heterogeneous) - **T**:
10	13	Slope tensor. The shape of slope can be smaller then first input X;	Slope tensor. The shape of slope can be smaller then first input X;
11	14	if so, its shape must be unidirectional broadcastable to X	if so, its shape must be unidirectional broadcastable to X
12	15
13	16	**Outputs**	**Outputs**
14	17
15	18	* **Y** (heterogeneous) - **T**:	* **Y** (heterogeneous) - **T**:
16	19	Output tensor (same size as X)	Output tensor (same size as X)
17	20
18	21	**Type Constraints**	**Type Constraints**
19	22
20	23	* **T** in (	* **T** in (
	24		tensor(bfloat16),
21	25	tensor(double),	tensor(double),
22	26	tensor(float),	tensor(float),
23	27	tensor(float16),	tensor(float16),
24	28	tensor(int32),	tensor(int32),
25	29	tensor(int64),	tensor(int64),
26	30	tensor(uint32),	tensor(uint32),
27	31	tensor(uint64)	tensor(uint64)
28	32	):	):
29	33	Constrain input and output types to float/int tensors.	Constrain input and output types to float/int tensors.

PRelu - 9

Version

• name: PRelu (GitHub)

• domain: main

• since_version: 9

• function: False

• support_level: SupportType.COMMON

• shape inference: True

This version of the operator has been available since version 9.

Summary

PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise. This operator supports unidirectional broadcasting (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX.

Inputs

• X (heterogeneous) - T: Input tensor

• slope (heterogeneous) - T: Slope tensor. The shape of slope can be smaller then first input X; if so, its shape must be unidirectional broadcastable to X

Outputs

• Y (heterogeneous) - T: Output tensor (same size as X)

Type Constraints

• T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.

Differences

0	0	PRelu takes input data (Tensor) and slope tensor as input, and produces one	PRelu takes input data (Tensor) and slope tensor as input, and produces one
1	1	output data (Tensor) where the function f(x) = slope * x for x < 0,	output data (Tensor) where the function f(x) = slope * x for x < 0,
2	2	f(x) = x for x >= 0., is applied to the data tensor elementwise.	f(x) = x for x >= 0., is applied to the data tensor elementwise.
3	3	This operator supports **unidirectional broadcasting** (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX _.	This operator supports **unidirectional broadcasting** (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX _.
4	4
5	5	**Inputs**	**Inputs**
6	6
7	7	* **X** (heterogeneous) - **T**:	* **X** (heterogeneous) - **T**:
8	8	Input tensor	Input tensor
9	9	* **slope** (heterogeneous) - **T**:	* **slope** (heterogeneous) - **T**:
10	10	Slope tensor. The shape of slope can be smaller then first input X;	Slope tensor. The shape of slope can be smaller then first input X;
11	11	if so, its shape must be unidirectional broadcastable to X	if so, its shape must be unidirectional broadcastable to X
12	12
13	13	**Outputs**	**Outputs**
14	14
15	15	* **Y** (heterogeneous) - **T**:	* **Y** (heterogeneous) - **T**:
16	16	Output tensor (same size as X)	Output tensor (same size as X)
17	17
18	18	**Type Constraints**	**Type Constraints**
19	19
20	20	* **T** in (	* **T** in (
21	21	tensor(double),	tensor(double),
22	22	tensor(float),	tensor(float),
23	23	tensor(float16)	tensor(float16),
	24		tensor(int32),
	25		tensor(int64),
	26		tensor(uint32),
	27		tensor(uint64)
24	28	):	):
25	29	Constrain input and output types to float tensors.	Constrain input and output types to float/int tensors.

PRelu - 7

Version

• name: PRelu (GitHub)

• domain: main

• since_version: 7

• function: False

• support_level: SupportType.COMMON

• shape inference: True

This version of the operator has been available since version 7.

Summary

PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise. This operator supports unidirectional broadcasting (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX.

Inputs

• X (heterogeneous) - T: Input tensor

• slope (heterogeneous) - T: Slope tensor. The shape of slope can be smaller then first input X; if so, its shape must be unidirectional broadcastable to X

Outputs

• Y (heterogeneous) - T: Output tensor (same size as X)

Type Constraints

• T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.

Differences

0	0	PRelu takes input data (Tensor) and slope tensor as input, and produces one	PRelu takes input data (Tensor) and slope tensor as input, and produces one
1	1	output data (Tensor) where the function f(x) = slope * x for x < 0,	output data (Tensor) where the function f(x) = slope * x for x < 0,
2	2	f(x) = x for x >= 0., is applied to the data tensor elementwise.	f(x) = x for x >= 0., is applied to the data tensor elementwise.
	3		This operator supports **unidirectional broadcasting** (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX _.
3	4
4	5	**Inputs**	**Inputs**
5	6
6	7	* **X** (heterogeneous) - **T**:	* **X** (heterogeneous) - **T**:
7	8	Input tensor	Input tensor
8	9	* **slope** (heterogeneous) - **T**:	* **slope** (heterogeneous) - **T**:
9	10	Slope tensor. If Slope is of size 1, the value is sharedacross	Slope tensor. The shape of slope can be smaller then first input X;
10	11	different channels	if so, its shape must be unidirectional broadcastable to X
11	12
12	13	**Outputs**	**Outputs**
13	14
14	15	* **Y** (heterogeneous) - **T**:	* **Y** (heterogeneous) - **T**:
15	16	Output tensor	Output tensor (same size as X)
16	17
17	18	**Type Constraints**	**Type Constraints**
18	19
19	20	* **T** in (	* **T** in (
20	21	tensor(double),	tensor(double),
21	22	tensor(float),	tensor(float),
22	23	tensor(float16)	tensor(float16)
23	24	):	):
24	25	Constrain input and output types to float tensors.	Constrain input and output types to float tensors.

PRelu - 6

Version

• name: PRelu (GitHub)

• domain: main

• since_version: 6

• function: False

• support_level: SupportType.COMMON

• shape inference: True

This version of the operator has been available since version 6.

Summary

PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise.

Inputs

• X (heterogeneous) - T: Input tensor

• slope (heterogeneous) - T: Slope tensor. If Slope is of size 1, the value is sharedacross different channels

Outputs

• Y (heterogeneous) - T: Output tensor

Type Constraints

• T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.

Differences

0	0	PRelu takes input data (Tensor) and slope tensor as input, and produces one	PRelu takes input data (Tensor) and slope tensor as input, and produces one
1	1	output data (Tensor) where the function f(x) = slope * x for x < 0,	output data (Tensor) where the function f(x) = slope * x for x < 0,
2	2	f(x) = x for x >= 0., is applied to the data tensor elementwise.	f(x) = x for x >= 0., is applied to the data tensor elementwise.
3	3
4		**Attributes**
5
6		* **consumed_inputs**:
7		legacy optimization attribute.
8
9	4	**Inputs**	**Inputs**
10	5
11	6	* **X** (heterogeneous) - **T**:	* **X** (heterogeneous) - **T**:
12	7	Input tensor	Input tensor
13	8	* **slope** (heterogeneous) - **T**:	* **slope** (heterogeneous) - **T**:
14	9	Slope tensor. If Slope is of size 1, the value is sharedacross	Slope tensor. If Slope is of size 1, the value is sharedacross
15	10	different channels	different channels
16	11
17	12	**Outputs**	**Outputs**
18	13
19	14	* **Y** (heterogeneous) - **T**:	* **Y** (heterogeneous) - **T**:
20	15	Output tensor	Output tensor
21	16
22	17	**Type Constraints**	**Type Constraints**
23	18
24	19	* **T** in (	* **T** in (
25	20	tensor(double),	tensor(double),
26	21	tensor(float),	tensor(float),
27	22	tensor(float16)	tensor(float16)
28	23	):	):
29	24	Constrain input and output types to float tensors.	Constrain input and output types to float tensors.

PRelu - 1

Version

• name: PRelu (GitHub)

• domain: main

• since_version: 1

• function: False

• support_level: SupportType.COMMON

• shape inference: False

This version of the operator has been available since version 1.

Summary

PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise.

Attributes

• consumed_inputs: legacy optimization attribute.

Inputs

• X (heterogeneous) - T: Input tensor

• slope (heterogeneous) - T: Slope tensor. If Slope is of size 1, the value is sharedacross different channels

Outputs

• Y (heterogeneous) - T: Output tensor

Type Constraints

• T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.