『TensorFlow』第四弹_classification分类学习_拨云见日

 1 import tensorflow as tf
 2 from tensorflow.examples.tutorials.mnist import input_data
 3 
 4 ‘‘‘数据下载‘‘‘
 5 # one_hot标签：[0,0...1,0...]
 6 mnist = input_data.read_data_sets(‘Mnist_data‘,one_hot=True)
 7 
 8 ‘‘‘生成层函数‘‘‘
 9 def add_layer(input,in_size,out_size,n_layer=‘layer‘,activation_function=None):
10     layer_name = ‘layer%s‘ % n_layer
11     with tf.name_scope(layer_name):
12         with tf.name_scope(‘weights‘):
13             Weights = tf.Variable(tf.random_normal([in_size,out_size]),name=‘W‘)
14             tf.summary.histogram(layer_name+‘/weights‘,Weights)
15         with tf.name_scope(‘biases‘):
16             biases = tf.Variable(tf.zeros([1,out_size]) + 0.1)
17             tf.summary.histogram(layer_name + ‘/biases‘, biases)
18         with tf.name_scope(‘Wx_plus_b‘):
19             # [in]*[[out]*in]+[out]
20             Wx_plus_b = tf.matmul(input,Weights) + biases
21         if activation_function is None:
22             outputs = Wx_plus_b
23         else:
24             outputs = activation_function(Wx_plus_b)
25         tf.summary.histogram(layer_name + ‘/outputs‘, outputs)
26         return outputs
27 
28 ‘‘‘准确率‘‘‘
29 def compute_accuracy(v_xs,v_ys):
30     global prediction
31     y_pre = sess.run(prediction,feed_dict={xs:v_xs}) #<------------------------
32     # tf.equal()对比预测值的索引和实际label的索引是否一样，一样返回True，不一样返回False
33     correct_prediction = tf.equal(tf.argmax(y_pre,1),tf.argmax(v_ys,1))
34     # 将pred即True或False转换为1或0,并对所有的判断结果求均值
35     accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
36     result = sess.run(accuracy,feed_dict={xs:v_xs,ys:v_ys}) #<------------------------
37     return result
38 
39 
40 ‘‘‘占位符‘‘‘
41 xs = tf.placeholder(tf.float32,[None,784])
42 ys = tf.placeholder(tf.float32,[None,10])
43 
44 ‘‘‘添加层‘‘‘
45 prediction = add_layer(xs,784,10,activation_function=tf.nn.softmax)  # softmax:分类用激活函数
46 
47 ‘‘‘计算loss‘‘‘
48 # 交叉熵损失函数，参数分别为预测值pred和实际label值y，reduce_mean为求batch平均loss
49 cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys*tf.log(prediction),reduction_indices=[1])) #<------------------------
50 train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy) # 梯度下降优化器
51 
52 ‘‘‘会话生成‘‘‘
53 with tf.Session() as sess:
54     sess.run(tf.global_variables_initializer())
55     for i in range(1000):
56         batch_xs,batch_ys = mnist.train.next_batch(100) # 逐个batch的去取数据
57         sess.run(train_step,feed_dict={xs:batch_xs,ys:batch_ys})
58         if i % 50 == 0:
59             print(compute_accuracy(mnist.test.images,mnist.test.labels))

1 ‘‘‘准确率‘‘‘
2 def compute_accuracy(v_xs,v_ys):
3     global prediction
4     y_pre = sess.run(prediction,feed_dict={xs:v_xs}) 
5     correct_prediction = tf.equal(tf.argmax(y_pre,1),tf.argmax(v_ys,1))
6     accuracy = tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
7     result = sess.run(accuracy,feed_dict={xs:v_xs,ys:v_ys}) 
8     print(y_pre,v_ys) #<------------------------
9     return result

1 def argmax(input, axis=None, name=None, dimension=None):