(17)tensorflow多层神经网络实现

多层神经网络实现

多层神经网络

张量方式实现

1. 初始化各层
2. 调用梯度记录器（自动求导）
3. 搭建各层

import tensorflow as tf
from tensorflow.keras import layers
x = tf.random.normal([3,784])#模拟2个样本，50个特征
w1 = tf.Variable(tf.random.truncated_normal([784,256],stddev=0.5))#第1层初始化
b1 = tf.zeros([256])
w2 = tf.Variable(tf.random.truncated_normal([256,128],stddev=0.5))#第2层初始化
b2 = tf.zeros([128])
w3 = tf.Variable(tf.random.truncated_normal([128,64],stddev=0.5))#第3层初始化
b3 = tf.zeros([64])
w4 = tf.Variable(tf.random.truncated_normal([64,10],stddev=0.5))#第4层初始化
b4 = tf.zeros([10])

with tf.GradientTape() as tape: # 梯度记录器
    o1 = tf.matmul(x,w1) + b1 #也可使用广播机制拓展b1, tf.broadcast_to(b1, [x.shape[0], 256])
    o1 = tf.nn.relu(o1)
    o2 = tf.matmul(o1,w2) + b2
    o2 = tf.nn.relu(o2)
    o3 =tf.matmul(o2,w3) + b3
    o3 = tf.nn.relu(o3)
    o4 =tf.matmul(o3,w4) + b4
    print(o4)

out:
tf.Tensor(
[[-184.37228   250.41742   790.16876  -711.15125  -268.63834   246.24713
  -434.43478   547.58203   591.3756   1196.3761  ]
 [-841.72205  -416.0834    166.62344  -616.033    -439.455      59.845196
   318.67456    60.858128  383.62747  1009.4866  ]
 [  -9.59348   297.99423  -370.27063  -913.0671    -73.28527   449.61548
  -586.05334  -529.05145   674.3345    610.58124 ]], shape=(3, 10), dtype=float32)

层方式实现

简单调用

import tensorflow as tf
from tensorflow.keras import layers
fc1 = layers.Dense(256,activation=tf.nn.relu)
fc2 = layers.Dense(128,activation=tf.nn.relu)
fc3 = layers.Dense(64,activation=tf.nn.relu)
fc4 = layers.Dense(10,activation=None)
x = tf.random.normal([3,256])
o1 = fc1(x)
o2 = fc2(o1)
o3 = fc3(o2)
o4 = fc4(o3)
print(o4)

out:
tf.Tensor(
[[-0.49023932 -0.04196656 -0.5316457  -0.22760229 -0.64782906  0.4603924
  -0.46672398  0.54325604 -0.7902754  -0.15966915]
 [-0.17537238  0.15008762 -0.3496042  -0.17603163  0.50083745 -0.62207043
   0.01685877  0.796759   -0.34656522  1.0465541 ]
 [-0.01338768 -0.16533731  0.11743313 -0.1506098  -0.65836793  0.3003255
   0.21182871  0.22682111  0.33577356  0.33805916]], shape=(3, 10), dtype=float32)

封装调用

• 使用Sequential作为容器

import tensorflow as tf
from tensorflow.keras import layers
from tensorflow.keras import Sequential
modle = Sequential([
    layers.Dense(256,activation=tf.nn.relu),
    layers.Dense(128,activation=tf.nn.relu),
    layers.Dense(64,activation=tf.nn.relu),
    layers.Dense(10,activation=None)])
x = tf.random.normal([3,256])
a = modle(x)
print(a)

out:
tf.Tensor(
[[-0.48306665  0.3618398  -0.27131498 -0.80744046 -0.6388896  -0.26316196
   0.8380004   0.341826    0.15942936 -0.19762628]
 [ 0.03107908  0.07116681 -0.49450177 -0.48394847 -0.40307134 -0.2677082
   0.4005356   0.29835856  0.0189686  -0.2250806 ]
 [ 0.55381507  0.6946548  -0.7538996  -1.324675   -0.59704345  0.39796948
   0.6455777   0.02428103 -0.31168345 -0.51364326]], shape=(3, 10), dtype=float32)