9优化器

# tensorflow 上的优化器
# 1.tf.train.GradientDescentOptimizer
# 2.tf.train.AdadeltaOptimizer
# 3.tf.train.AdagradOptimizer
# 4.tf.train.AdagradDAOptimizer
# 5.tf.train.MomentumOptimizer
# 6.tf.train.AdamOptimizer
# 7.tf.train.FtrlOptimizer
# 8.tf.train.ProximalGradientDescentOptimizer
# 9.tf.train.ProximalAdagradOptimizer
# 10.tf.train.RMSPropOptimizer

# 各种优化器对比:链接

#最常用的是AdamOptimizer，这里就通过adma和衰减的学习率加上之前的多层结构，使手写数字模型准确率达到98%以上

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
 
#载入数据集
mnist=input_data.read_data_sets("MNIST_data", one_hot=True)
 
#每个批次的大小
batch_size=100
#计算一共有多少批次
n_batch=mnist.train.num_examples // batch_size
 
#定义两个placeholder
x=tf.placeholder(tf.float32,[None,784])
y=tf.placeholder(tf.float32,[None,10])
keep_prob=tf.placeholder(tf.float32)
lr=tf.Variable(0.001, dtype=tf.float32)#学习率变量
 
#创建一个简单的神经网络
W1=tf.Variable(tf.truncated_normal([784,200],stddev=0.1))   #这里我们使用一个截断的正太分布初始化W
b1=tf.Variable(tf.zeros([1,200]))
L1=tf.nn.tanh(tf.matmul(x,W1)+b1)   #激活函数为双曲正切函数
L1_drop=tf.nn.dropout(L1, keep_prob)
 
W2=tf.Variable(tf.truncated_normal([200,100],stddev=0.1))
b2=tf.Variable(tf.zeros([1,100]))
L2=tf.nn.tanh(tf.matmul(L1_drop, W2)+b2)
L2_drop=tf.nn.dropout(L2, keep_prob)
 
W3=tf.Variable(tf.truncated_normal([100,10], stddev=0.1))
b3=tf.Variable(tf.zeros([1,10]))
prediction=tf.nn.softmax(tf.matmul(L2_drop,W3)+b3)
 
#交叉熵代价函数
loss=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=prediction))
#使用Adam优化器
train_step=tf.train.AdamOptimizer(lr).minimize(loss)
 
#初始化变量
init=tf.global_variables_initializer()
 
#结果存放在一个布尔型列表中
correct_prediction=tf.equal(tf.argmax(y,1), tf.argmax(prediction,1)) #argmax返回一维张量中最大的值所在的位置
#求准确率
accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
 
with tf.Session() as sess:
    sess.run(init)
    for epoch in range(51):
        sess.run(tf.assign(lr, 0.001*(0.95**epoch)))#每训练一次改变学习率，学习率逐渐减小（开始快速收敛，最后逐渐降低）
        for batch in range(n_batch):
            batch_xs,batch_ys=mnist.train.next_batch(batch_size)
            sess.run(train_step,feed_dict={x:batch_xs, y:batch_ys,keep_prob:1})
 
        test_acc=sess.run(accuracy,feed_dict={x:mnist.test.images, y:mnist.test.labels,keep_prob:1.0})
        train_acc=sess.run(accuracy,feed_dict={x:mnist.train.images, y:mnist.train.labels,keep_prob:1.0})
        learing_rate=sess.run(lr)
        print("Iter"+str(epoch)+",Testing Accuracy "+str(test_acc)+"Training Accuracy "+str(train_acc)
              +",learing_rate "+str(learing_rate))

Extracting MNIST_data\train-images-idx3-ubyte.gz
Extracting MNIST_data\train-labels-idx1-ubyte.gz
Extracting MNIST_data\t10k-images-idx3-ubyte.gz
Extracting MNIST_data\t10k-labels-idx1-ubyte.gz
Iter0,Testing Accuracy 0.9398Training Accuracy 0.94338185,learing_rate 0.001
Iter1,Testing Accuracy 0.9551Training Accuracy 0.9588,learing_rate 0.00095
Iter2,Testing Accuracy 0.964Training Accuracy 0.97063637,learing_rate 0.0009025
Iter3,Testing Accuracy 0.9684Training Accuracy 0.97656363,learing_rate 0.000857375
Iter4,Testing Accuracy 0.9702Training Accuracy 0.9806,learing_rate 0.00081450626
Iter5,Testing Accuracy 0.9704Training Accuracy 0.9821636,learing_rate 0.0007737809
Iter6,Testing Accuracy 0.9715Training Accuracy 0.9852545,learing_rate 0.0007350919
Iter7,Testing Accuracy 0.972Training Accuracy 0.98676366,learing_rate 0.0006983373
Iter8,Testing Accuracy 0.973Training Accuracy 0.9878909,learing_rate 0.0006634204
Iter9,Testing Accuracy 0.9745Training Accuracy 0.99023634,learing_rate 0.0006302494
Iter10,Testing Accuracy 0.9742Training Accuracy 0.99012727,learing_rate 0.0005987369
Iter11,Testing Accuracy 0.9753Training Accuracy 0.99194545,learing_rate 0.0005688001
Iter12,Testing Accuracy 0.976Training Accuracy 0.9920909,learing_rate 0.0005403601
Iter13,Testing Accuracy 0.9765Training Accuracy 0.9929818,learing_rate 0.0005133421
Iter14,Testing Accuracy 0.976Training Accuracy 0.9932,learing_rate 0.000487675
Iter15,Testing Accuracy 0.9757Training Accuracy 0.9932182,learing_rate 0.00046329122
Iter16,Testing Accuracy 0.9768Training Accuracy 0.99356365,learing_rate 0.00044012666
Iter17,Testing Accuracy 0.9763Training Accuracy 0.9940909,learing_rate 0.00041812033
Iter18,Testing Accuracy 0.978Training Accuracy 0.9944909,learing_rate 0.00039721432
Iter19,Testing Accuracy 0.978Training Accuracy 0.99456364,learing_rate 0.0003773536
Iter20,Testing Accuracy 0.978Training Accuracy 0.99485457,learing_rate 0.00035848594
Iter21,Testing Accuracy 0.9779Training Accuracy 0.99496365,learing_rate 0.00034056162
Iter22,Testing Accuracy 0.9775Training Accuracy 0.99505454,learing_rate 0.00032353355
Iter23,Testing Accuracy 0.9784Training Accuracy 0.9951818,learing_rate 0.00030735688
Iter24,Testing Accuracy 0.9788Training Accuracy 0.9953455,learing_rate 0.000291989
Iter25,Testing Accuracy 0.9795Training Accuracy 0.9954,learing_rate 0.00027738957
Iter26,Testing Accuracy 0.9794Training Accuracy 0.9954727,learing_rate 0.0002635201
Iter27,Testing Accuracy 0.9774Training Accuracy 0.9953455,learing_rate 0.00025034408
Iter28,Testing Accuracy 0.9786Training Accuracy 0.9954727,learing_rate 0.00023782688
Iter29,Testing Accuracy 0.9787Training Accuracy 0.9954909,learing_rate 0.00022593554
Iter30,Testing Accuracy 0.9787Training Accuracy 0.9955818,learing_rate 0.00021463877
Iter31,Testing Accuracy 0.9789Training Accuracy 0.99561816,learing_rate 0.00020390682
Iter32,Testing Accuracy 0.979Training Accuracy 0.99570906,learing_rate 0.00019371149
Iter33,Testing Accuracy 0.9786Training Accuracy 0.9957455,learing_rate 0.0001840259
Iter34,Testing Accuracy 0.9788Training Accuracy 0.9958,learing_rate 0.00017482461
Iter35,Testing Accuracy 0.979Training Accuracy 0.9958,learing_rate 0.00016608338
Iter36,Testing Accuracy 0.979Training Accuracy 0.9958182,learing_rate 0.00015777921
Iter37,Testing Accuracy 0.9787Training Accuracy 0.9958909,learing_rate 0.00014989026
Iter38,Testing Accuracy 0.9785Training Accuracy 0.9959091,learing_rate 0.00014239574
Iter39,Testing Accuracy 0.9791Training Accuracy 0.99596363,learing_rate 0.00013527596
Iter40,Testing Accuracy 0.979Training Accuracy 0.99596363,learing_rate 0.00012851215
Iter41,Testing Accuracy 0.9793Training Accuracy 0.9960182,learing_rate 0.00012208655
Iter42,Testing Accuracy 0.9782Training Accuracy 0.99605453,learing_rate 0.00011598222
Iter43,Testing Accuracy 0.9785Training Accuracy 0.99603635,learing_rate 0.00011018311
Iter44,Testing Accuracy 0.9778Training Accuracy 0.99603635,learing_rate 0.000104673956
Iter45,Testing Accuracy 0.9788Training Accuracy 0.9960909,learing_rate 9.944026e-05
Iter46,Testing Accuracy 0.979Training Accuracy 0.99610907,learing_rate 9.446825e-05
Iter47,Testing Accuracy 0.9793Training Accuracy 0.99612725,learing_rate 8.974483e-05
Iter48,Testing Accuracy 0.9786Training Accuracy 0.99612725,learing_rate 8.525759e-05
Iter49,Testing Accuracy 0.9789Training Accuracy 0.9961636,learing_rate 8.099471e-05
Iter50,Testing Accuracy 0.9788Training Accuracy 0.9962182,learing_rate 7.6944976e-05