Tensorflow框架学习(1)--CNN的实现

# CNN神经网络的代码实现
import tensorflow as tf
# 导入数据集
from tensorflow.examples.tutorials.mnist import input_data
# number 1 to 10 data:读取数据
mnist = input_data.read_data_sets("MNIST_data", one_hot=True)

sess = tf.Session()

#计算准确率
def compute_accuracy(v_xs, v_ys):
    global prediction
    y_pre = sess.run(prediction, feed_dict={xs: v_xs, ys: v_ys, keep_prob: 0.5})   # keep_prob的设置是为了防止发生过拟合的现象
    correct_prediction = tf.equal(tf.argmax(y_pre, 1), tf.argmax(v_ys, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
    result = sess.run(accuracy, feed_dict={xs: v_xs, ys: v_ys, keep_prob: 0.5})
    return result

# 定义并初始化weight_variable
def weight_variable(shape):
    initial = tf.truncated_normal(shape, stddev=0.1)
    return tf.Variable(initial)

# 定义并初始化bias_variable
def bias_variable(shape):
    initial = tf.constant(0.1, shape=shape)
    return tf.Variable(initial)

#定义卷积层
def conv2d(x,W):
    # strides=[1,x_movement,y_movement,1]
    # strides[0] = strides[3] = 0
    return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding="SAME")

#定义池化层
def max_pool_2x2(x):
    # strides=[1,x_movement,y_movement,1]
    # strides[0] = strides[3] = 0
    # ksize=[1,x,x,1]:其中表示池化的大小
    # 在池化得我过程中，不能有重叠，否则会出现报错 --栽沟里了！！！
    return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="SAME")

# 定义placeholder for inputs to network,None表示输入的样本数待定
xs = tf.placeholder(tf.float32, [None, 784])     # None表示输入的样本数待定（待定）
ys = tf.placeholder(tf.float32, [None, 10])
keep_prob = tf.placeholder(tf.float32)

# 将输入的数据变换成矩阵的形式
# 此时x_image表示的是[n_samples,28,28,1]，其中1表示channel
x_image = tf.reshape(xs, [-1, 28, 28, 1])        # -1表示输入的样本数待定（待定）

# 定义conv1 layer--卷积层1
# [5,5,1,32]
# 5*5:patch size  1:channel  32:filter的数量
W_conv1 = weight_variable([5, 5, 1, 32])
b_conv1 = bias_variable([32])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)   #output_size:28*28*32
h_pool1 = max_pool_2x2(h_conv1)                           #output_size:14*14*32

# 定义conv2 layer
# 32：第二层卷积中卷积核的深度   64：卷积核的个数
W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)   #output_size:14*14*64
h_pool2 = max_pool_2x2(h_conv2)                           #output_size:7*7*64


# 定义func1 layer:全卷积层1
W_fc1 = weight_variable([7*7*64, 1024])    #全连接层的in_size:7*7*64  out_size:1024
b_fc1 = bias_variable([1024])
# 将con2 layer的输出拉平:[n_sample,7,7,64] ->> [n_sample,7*7*64]
h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])        # -1表示输入的样本数（待定）
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
# 为了防止出现过拟合，使用tf.nn.dropout()
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

# 定义func2 layer
W_fc2 = weight_variable([1024, 10])    #全连接层的in_size:1024  out_size:10(分类对应10个数据)
b_fc2 = bias_variable([10])
# 将con2 layer的输出拉平:[n_sample,7,7,64] ->> [n_sample,7*7*64]
prediction = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2)+b_fc2)

# the error between prediction and real data
cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys * tf.log(prediction), reduction_indices=[1]))
# 优化器的选择
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)

# 定义变量需要初始化所有的参数  -- very important
sess.run(tf.initialize_all_variables())

# 数据的训练
for i in range(1000):
    batch_xs, batch_ys = mnist.train.next_batch(100)
    sess.run(train_step, feed_dict={xs: batch_xs, ys: batch_ys, keep_prob: 0.5})
    if i % 50 == 0:
        #    每训练50次，计算一下准确略：函数需要自定义
        print(compute_accuracy(
            mnist.test.images, mnist.test.labels))

说明：

　　①在实现CNN过程中，定义池化层的时候，由于strides设置错误（[1, 1, 1, 1] ->>[1, 2, 2, 1]），导致在进行MaxPooling时出现重叠进而导致后续的运算出现矩阵维度不匹配的错误。

　　②以上代码实在tensorflow1.13.0环境中运行的--采用的是符号式编程。