1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
|
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT License.
# Features included:
# Overloaded operators such as <=, +, /
# Nested expressions
# Type annotation for attributes (to distinguish inputs and attributes)
# Importing (predefined) opsets
# Features not yet covered:
# Auto-cast for constants/attributes. (Must include explicit CastLike ops)
# No default-values for attributes yet, which are currently present in the opschema definition (in ONNX).
# No default-values for inputs yet.
# Element-type annotation for tensors
from onnxscript import script
from onnxscript.onnx_opset import opset18 as op
@script()
def Relu(X):
zero = op.CastLike(0, X)
return op.Max(X, zero)
@script()
def Selu(X, alpha: float, gamma: float):
alphaX = op.CastLike(alpha, X)
gammaX = op.CastLike(gamma, X)
neg = gammaX * (alphaX * op.Exp(X) - alphaX)
pos = gammaX * X
zero = op.CastLike(0, X)
return op.Where(X <= zero, neg, pos)
@script()
def Elu(X, alpha: float):
alphaX = op.CastLike(alpha, X)
zero = op.CastLike(0, X)
one = op.CastLike(1, X)
return op.Where(X < zero, alphaX * (op.Exp(X) - one), X)
@script()
def ThresholdedRelu(X, alpha: float):
zero = op.CastLike(0, X)
alphaX = op.CastLike(alpha, X)
return op.Where(X > alphaX, X, zero)
@script()
def LeakyRelu(X, alpha: float):
zero = op.CastLike(0, X)
alphaX = op.CastLike(alpha, X)
return op.Where(X < zero, alphaX * X, X)
@script()
def PRelu(X, slope):
zero = op.CastLike(0, X)
return op.Where(X < zero, slope * X, X)
@script()
def HardSigmoid(X, alpha: float, beta: float):
zero = op.CastLike(0, X)
one = op.CastLike(1, X)
alphaX = op.CastLike(alpha, X)
betaX = op.CastLike(beta, X)
return op.Max(zero, op.Min(one, alphaX * X + betaX))
@script()
def Shrink(x, bias: float, lambd: float):
zero = op.CastLike(0, x)
return op.Where(x < -lambd, x + bias, op.Where(x > lambd, x - bias, zero))
@script()
def Softplus(X):
one = op.CastLike(1, X)
return op.Log(op.Exp(X) + one)
@script()
def Softsign(X):
one = op.CastLike(1, X)
return X / (one + op.Abs(X))
from onnxscript.onnx_types import BOOL, FLOAT
@script()
def Clip(input: FLOAT[...], min: FLOAT = None, max: FLOAT = None) -> FLOAT[...]:
result = input
if op.OptionalHasElement(min):
result = op.Where(result < min, min, result)
if op.OptionalHasElement(max):
result = op.Where(result > max, max, result)
return result
# @script()
# def Clip(input, min, max):
# return op.Where(input < min, min, op.Where(input > max, max, input))
@script()
def OptionalHasElement(input: FLOAT[...]) -> BOOL:
return op.OptionalHasElement(input)
|
