input_signature的好處:
1.可以限定函數(shù)的輸入類型,以防止調(diào)用函數(shù)時(shí)調(diào)錯(cuò),
2.一個(gè)函數(shù)有了input_signature之后,在tensorflow里邊才可以保存成savedmodel。在保存成savedmodel的過程中,需要使用get_concrete_function函數(shù)把一個(gè)tf.function標(biāo)注的普通的python函數(shù)變成帶有圖定義的函數(shù)。
下面的代碼具體體現(xiàn)了input_signature可以限定函數(shù)的輸入類型這一作用。
1
2
3
4
5
6
7
8
|
@tf .function(input_signature = [tf.TensorSpec([ None ], tf.int32, name = 'x' )]) def cube(z): #實(shí)現(xiàn)輸入的立方 return tf. pow (z, 3 ) try : print (cube(tf.constant([ 1. , 2. , 3. ]))) except ValueError as ex: print (ex) print (cube(tf.constant([ 1 , 2 , 3 ]))) |
輸出:
Python inputs incompatible with input_signature:
inputs: (
tf.Tensor([1. 2. 3.], shape=(3,), dtype=float32))
input_signature: (
TensorSpec(shape=(None,), dtype=tf.int32, name='x'))
tf.Tensor([ 1 8 27], shape=(3,), dtype=int32)
get_concrete_function的使用
note:首先說明,下面介紹的函數(shù)在模型構(gòu)建、模型訓(xùn)練的過程中不會(huì)用到,下面介紹的函數(shù)主要用在兩個(gè)地方:1、如何保存模型 2、保存好模型后,如何載入進(jìn)來。
可以給 由@tf.function標(biāo)注的普通的python函數(shù),給它加上input_signature, 從而讓這個(gè)python函數(shù)變成一個(gè)可以保存的tensorflow圖結(jié)構(gòu)(SavedModel)
舉例說明函數(shù)的用法:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
|
@tf .function(input_signature = [tf.TensorSpec([ None ], tf.int32, name = 'x' )]) def cube(z): return tf. pow (z, 3 ) try : print (cube(tf.constant([ 1. , 2. , 3. ]))) except ValueError as ex: print (ex) print (cube(tf.constant([ 1 , 2 , 3 ]))) # @tf.function py func -> tf graph # get_concrete_function -> add input signature -> SavedModel cube_func_int32 = cube.get_concrete_function( tf.TensorSpec([ None ], tf.int32)) #tensorflow的類型 print (cube_func_int32) |
輸出:
<tensorflow.python.eager.function.ConcreteFunction object at 0x00000240E29695C0>
從輸出結(jié)果可以看到:調(diào)用get_concrete_function函數(shù)后,輸出的是一個(gè)ConcreteFunction對(duì)象
1
2
3
4
5
|
#看用新參數(shù)獲得的對(duì)象與原來的對(duì)象是否一樣 print (cube_func_int32 is cube.get_concrete_function( tf.TensorSpec([ 5 ], tf.int32))) #輸入大小為5 print (cube_func_int32 is cube.get_concrete_function( tf.constant([ 1 , 2 , 3 ]))) #傳具體數(shù)據(jù) |
輸出:
True
True
cube_func_int32.graph #圖定義
輸出:
1
2
3
4
|
[<tf.Operation 'x' type = Placeholder>, <tf.Operation 'Pow/y' type = Const>, <tf.Operation 'Pow' type = Pow >, <tf.Operation 'Identity' type = Identity>] |
1
2
|
pow_op = cube_func_int32.graph.get_operations()[ 2 ] print (pow_op) |
輸出:
name: "Pow"
op: "Pow"
input: "x"
input: "Pow/y"
attr {
key: "T"
value {
type: DT_INT32
}
}
1
2
|
print ( list (pow_op.inputs)) print ( list (pow_op.outputs)) |
輸出:
[<tf.Tensor 'x:0' shape=(None,) dtype=int32>, <tf.Tensor 'Pow/y:0' shape=() dtype=int32>]
[<tf.Tensor 'Pow:0' shape=(None,) dtype=int32>]
cube_func_int32.graph.get_operation_by_name("x")
輸出:
<tf.Operation 'x' type=Placeholder>
cube_func_int32.graph.get_tensor_by_name("x:0") #默認(rèn)加“:0”
<tf.Tensor 'x:0' shape=(None,) dtype=int32>
cube_func_int32.graph.as_graph_def() #總名字,針對(duì)上面兩個(gè)
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
|
node { name: "x" op: "Placeholder" attr { key: "_user_specified_name" value { s: "x" } } attr { key: "dtype" value { type : DT_INT32 } } attr { key: "shape" value { shape { dim { size: - 1 } } } } } node { name: "Pow/y" op: "Const" attr { key: "dtype" value { type : DT_INT32 } } attr { key: "value" value { tensor { dtype: DT_INT32 tensor_shape { } int_val: 3 } } } } node { name: "Pow" op: "Pow" input : "x" input : "Pow/y" attr { key: "T" value { type : DT_INT32 } } } node { name: "Identity" op: "Identity" input : "Pow" attr { key: "T" value { type : DT_INT32 } } } versions { producer: 119 } |
到此這篇關(guān)于tensorflow2.0的函數(shù)簽名與圖結(jié)構(gòu)的文章就介紹到這了,更多相關(guān)tensorflow函數(shù)簽名與圖結(jié)構(gòu)內(nèi)容請(qǐng)搜索服務(wù)器之家以前的文章或繼續(xù)瀏覽下面的相關(guān)文章希望大家以后多多支持服務(wù)器之家!
原文鏈接:https://blog.csdn.net/qq_41660119/article/details/105793777?