tensorflow (4) mnist·CNN

0.文件目錄如圖

  

1.CNN實現如下，每行代碼都有註釋

import tensorflow as tf
import input_data
#卷積網絡
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)   #載入數據
sess = tf.InteractiveSession()  #創建會話

x = tf.placeholder("float", shape=[None, 784]) #創建變量x,訓練時輸入數據，None一次輸入的樣本數量，784是照片轉換的數組
y_ = tf.placeholder("float", shape=[None, 10]) #創建變量y_，訓練時輸出數據  ，有十個數字，對應輸出數組長度爲10

def weight_variable(shape): #權重，shape待輸入
  initial = tf.truncated_normal(shape, stddev=0.1)   #tf.truncated_normal(shape, stddev=0.1),shape表示生成張量的維度，mean是均值，stddev是標準差，truncated爲截斷正態分佈
                                                     #生成[n,n]張量，一個方差爲零的正態分佈，這樣防止梯度爲零，應爲激勵函數爲RELU函數，初始權重爲零的話，節點輸出均爲零
  return tf.Variable(initial)    #返回權重變量初始值

def bias_variable(shape):  #偏置
  initial = tf.constant(0.1, shape=shape)  #偏置爲常量
  return tf.Variable(initial)


def conv2d(x, W):   ####卷積
  return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')  #[1，stride，stride，1]圖像只有兩維，中間兩個參數是二維方向上的步長 ‘VALID’ 'SAME'（卷積核可以停留在圖像邊緣）

def max_pool_2x2(x):  ####池化
  return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding='SAME')  #ksize：池化窗口的大小，取一個四維向量，一般是[1, height, width, 1]，因爲我們不想在batch和channels上做池化，所以這兩個維度設爲了1

###第一層卷積
W_conv1 = weight_variable([5, 5, 1, 32])  #參數爲shape，5*5，單通道輸入，32通道特徵輸出
b_conv1 = bias_variable([32])  #參數爲shape，32通道特徵輸出，故b的shape爲長度32  [ 0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1 0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1  0.1 0.1  0.1]


x_image = tf.reshape(x, [-1,28,28,1])  #[-1,28,28,1] batch數量不限，28*28像素的圖片  深度爲1
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)  #激活函數
h_pool1 = max_pool_2x2(h_conv1)                    #池化

###第二層卷積
W_conv2 = weight_variable([5, 5, 32, 64])   #5*5核，32輸入，64輸出
b_conv2 = bias_variable([64])

h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
h_pool2 = max_pool_2x2(h_conv2)

###密集連接層
W_fc1 = weight_variable([7 * 7 * 64, 1024])
b_fc1 = bias_variable([1024])

h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])   #把最後一層池化層轉爲  7*7*64列的矩陣
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)

###Dropout
keep_prob = tf.placeholder("float")
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

###輸出層
W_fc2 = weight_variable([1024, 10])
b_fc2 = bias_variable([10])

###訓練和評估模型
y_conv=tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)
cross_entropy = -tf.reduce_sum(y_*tf.log(y_conv))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
correct_prediction = tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
sess.run(tf.initialize_all_variables())
for i in range(20000):
  batch = mnist.train.next_batch(50)
  if i%100 == 0:
    train_accuracy = accuracy.eval(feed_dict={x:batch[0], y_: batch[1], keep_prob: 1.0})
    print("step %d, training accuracy %g"%(i, train_accuracy))
    # print(sess.run(y))
  train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})
print("test accuracy %g"%accuracy.eval(feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0}))