用CNN进行训练可以提高准确度,例如之前只是单独的加一个隐藏层,识别数字的准确度在87左右,但是用CNN在训练1000次的话,准确度能到达97左右。
本次运用mnist数据集,识别图片中的数字为0-9哪一个,有两个卷积层,两个全连接层,激活函数用的relu,最后softmax分类,训练1000次计算准确度并输出(每50次输出一次)。
from __future__ import print_function
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
# number 1 to 10 data
mnist = input_data.read_data_sets('E:/mnist_data', one_hot=True)
def compute_accuracy(v_xs, v_ys):
global prediction
y_pre = sess.run(prediction, feed_dict={xs: v_xs, keep_prob: 1})
correct_prediction = tf.equal(tf.argmax(y_pre, 1), tf.argmax(v_ys, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) #tf.cast将correct_prediction格式转换为float32
result = sess.run(accuracy, feed_dict={xs: v_xs, ys: v_ys, keep_prob: 1})
return result
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev=0.1) #产生一个随机权重,标准差为0.1
return tf.Variable(initial)
def bias_variable(shape):
initial = tf.constant(0.1, shape=shape) #产生bias,初始值为0.1
return tf.Variable(initial)
def conv2d(x, W): #定义卷积层
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME') # x即输入图片,W即权重, 步长,第1,4位必须为1, 第2,3位为x和y轴上的步长,padding为same时,卷积出来的东西和之前一样大
def max_pool_2x2(x): #pooling层,保留更多的图片信息
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
#define placeholder for inputs to network
xs = tf.placeholder(tf.float32, [None, 784])/255. #28*28
ys = tf.placeholder(tf.float32, [None, 10])
keep_prob = tf.placeholder(tf.float32)
x_image = tf.reshape(xs, [-1, 28, 28, 1]) #重定义输入,-1即有多少个sample都不管,然后28*28,然后黑白颜色为1,若彩色则为3
## conv1 layer ##
W_conv1 = weight_variable([5, 5, 1, 32]) #patch 5*5, in size 1, out size 32
b_conv1 = bias_variable([32])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1) #进行卷积,并relu激励 output size 28*28*32
h_pool1 = max_pool_2x2(h_conv1) #池化 output size 14*14*32
## conv2 layer ##
W_conv2 = weight_variable([5, 5, 32, 64]) #patch 5*5, in size 32, out size 64
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2) #进行卷积,并relu激励 output size 14*14*32
h_pool2 = max_pool_2x2(h_conv2) #池化 output size 7*7*32
## fc1 layer 全连接层 ##
W_fc1 = weight_variable([7*7*64, 1024]) #权重为这些行,1024列
b_fc1 = bias_variable([1024]) #b为1024列
h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64]) #把h_pool2转换为一列
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1) # wx+b 然后relu激活
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob) #防止overfitting, dropout一下
## fc2 layer ##
W_fc2 = weight_variable([1024, 10])
b_fc2 = bias_variable([10])
prediction = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2) #softmax生成一些判断的值,10列,判断数字到底是几
cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys*tf.log(prediction), reduction_indices=[1]))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
sess = tf.Session()
#sess.run(tf.initialize_all_variables())
if int((tf.__version__).split('.')[1]) < 12 and int((tf.__version__).split('.')[0]) < 1:
init = tf.initialize_all_variables()
else:
init = tf.global_variables_initializer()
sess.run(init)
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:
print(compute_accuracy(mnist.test.images[:1000], mnist.test.labels[:1000]))