全连接的层实现
功能函数代码
层实现方式layers.Dense(units, activation)获取 Dense 类的权值矩阵fc.kernel获取 Dense 类的偏置向量fc.bias返回待优化参数列表fc.trainable_variables
层实现
layers.Dense(units, activation)units指定层输出节点数activation指定激活函数ayers.Dense在调用时会根据输入数据自动生成输入节点数
import tensorflow
as tf
from tensorflow
.keras
import layers
x
= tf
.random
.normal
([3,50])
fc
= layers
.Dense
(2,activation
=tf
.nn
.relu
)
h1
= fc
(x
)
print('h1',h1
)
print('K',fc
.kernel
)
print('b',fc
.bias
)
print('v',fc
.trainable_variables
)
out
:
C
:\Users\Admin\PycharmProjects\untitled1\venv\Scripts\python
.exe C
:/Users
/Admin
/PycharmProjects
/untitled1
/tensor练习使用
.py
2020-09-04 13:06:21.911380: I tensorflow
/core
/platform
/cpu_feature_guard
.cc
:142] Your CPU supports instructions that this TensorFlow binary was
not compiled to use
: AVX2
h1 tf
.Tensor
(
[[0.95606685 1.1911137 ]
[0. 0. ]
[0.14124475 0.04009145]], shape
=(3, 2), dtype
=float32
)
K
<tf
.Variable
'dense/kernel:0' shape
=(50, 2) dtype
=float32
, numpy
=
array
([[ 0.0455344 , -0.19482273],
[-0.26869717, 0.02412117],
[ 0.21953332, -0.03901473],
[ 0.22162515, -0.32377893],
[ 0.30170476, 0.07104701],
[-0.24665953, 0.214688 ],
[ 0.11439314, 0.2840066 ],
[ 0.27506483, 0.2678889 ],
[-0.3137263 , 0.16774249],
[-0.08592525, 0.09441769],
[-0.21467607, -0.00248331],
[-0.23901463, 0.1094763 ],
[ 0.08673692, 0.26048535],
[-0.06203741, -0.1819801 ],
[ 0.18807596, 0.2989127 ],
[-0.25750422, 0.153965 ],
[ 0.12020397, 0.19332612],
[ 0.25078076, -0.12966038],
[ 0.06809869, 0.265687 ],
[-0.3342834 , -0.24474217],
[ 0.1052615 , -0.11368768],
[-0.12656587, 0.08862066],
[ 0.11732206, 0.3000934 ],
[ 0.3307876 , -0.3007087 ],
[ 0.14409512, -0.21482137],
[-0.04636192, 0.08430123],
[ 0.14868012, 0.26966405],
[-0.13744491, 0.18106598],
[ 0.05841842, -0.14975952],
[ 0.10092789, 0.2506609 ],
[ 0.31394488, 0.10897231],
[-0.22760901, -0.10806008],
[ 0.23610026, 0.2924561 ],
[ 0.19984013, -0.12296666],
[-0.21468846, -0.29082647],
[ 0.20162839, -0.24724546],
[-0.05970627, 0.33641344],
[ 0.30267793, -0.18490817],
[ 0.31117034, -0.31890184],
[ 0.09414282, 0.08582264],
[ 0.00278926, -0.08124155],
[ 0.17134207, 0.32664698],
[ 0.2573135 , 0.1359677 ],
[-0.31898403, 0.02981722],
[-0.29750133, -0.33452296],
[ 0.20652354, 0.09661892],
[-0.21122393, -0.265467 ],
[ 0.25332725, -0.15709157],
[ 0.31241155, -0.11074884],
[ 0.02461901, 0.18401676]], dtype
=float32
)>
b
<tf
.Variable
'dense/bias:0' shape
=(2,) dtype
=float32
, numpy
=array
([0., 0.], dtype
=float32
)>
v
[<tf
.Variable
'dense/kernel:0' shape
=(50, 2) dtype
=float32
, numpy
=
array
([[ 0.0455344 , -0.19482273],
[-0.26869717, 0.02412117],
[ 0.21953332, -0.03901473],
[ 0.22162515, -0.32377893],
[ 0.30170476, 0.07104701],
[-0.24665953, 0.214688 ],
[ 0.11439314, 0.2840066 ],
[ 0.27506483, 0.2678889 ],
[-0.3137263 , 0.16774249],
[-0.08592525, 0.09441769],
[-0.21467607, -0.00248331],
[-0.23901463, 0.1094763 ],
[ 0.08673692, 0.26048535],
[-0.06203741, -0.1819801 ],
[ 0.18807596, 0.2989127 ],
[-0.25750422, 0.153965 ],
[ 0.12020397, 0.19332612],
[ 0.25078076, -0.12966038],
[ 0.06809869, 0.265687 ],
[-0.3342834 , -0.24474217],
[ 0.1052615 , -0.11368768],
[-0.12656587, 0.08862066],
[ 0.11732206, 0.3000934 ],
[ 0.3307876 , -0.3007087 ],
[ 0.14409512, -0.21482137],
[-0.04636192, 0.08430123],
[ 0.14868012, 0.26966405],
[-0.13744491, 0.18106598],
[ 0.05841842, -0.14975952],
[ 0.10092789, 0.2506609 ],
[ 0.31394488, 0.10897231],
[-0.22760901, -0.10806008],
[ 0.23610026, 0.2924561 ],
[ 0.19984013, -0.12296666],
[-0.21468846, -0.29082647],
[ 0.20162839, -0.24724546],
[-0.05970627, 0.33641344],
[ 0.30267793, -0.18490817],
[ 0.31117034, -0.31890184],
[ 0.09414282, 0.08582264],
[ 0.00278926, -0.08124155],
[ 0.17134207, 0.32664698],
[ 0.2573135 , 0.1359677 ],
[-0.31898403, 0.02981722],
[-0.29750133, -0.33452296],
[ 0.20652354, 0.09661892],
[-0.21122393, -0.265467 ],
[ 0.25332725, -0.15709157],
[ 0.31241155, -0.11074884],
[ 0.02461901, 0.18401676]], dtype
=float32
)>, <tf
.Variable
'dense/bias:0' shape
=(2,) dtype
=float32
, numpy
=array
([0., 0.], dtype
=float32
)>]
