# coding=utf-8
from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
mnist = input_data.read_data_sets('/home/star/MNIST_data', one_hot=True)
sess = tf.InteractiveSession()
#give a normal noise to breake symmetry
def weigth_varible(shape):
initial = tf.truncated_normal(shape, stddev=0.1)
return tf.Variable(initial)
#give some biases to avoid dead neurons
def bias_varible(shape):
initial = tf.constant(0.1,shape=shape)
return tf.Variable(initial)
#strides:A list, move step,[batch, height, width, channels],when batch=1 then do not skip any batch, when channels=1 then do not skip any channel
#padding: A string, either `'VALID'` or `'SAME'
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides=[1,1,1,1], padding='SAME')
#strides:A list, move step,[batch, height, width, channels],when batch=1 then do not skip any batch, when channels=1 then do not skip any channel
#ksize: A list, kernel size, it's list has the same meaning as strides'
#padding: A string, either `'VALID'` or `'SAME'
def max_pool_2x2(x):
return tf.nn.max_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')
x = tf.placeholder(tf.float32, [None,784])
y_ = tf.placeholder(tf.float32, [None, 10])
#reshape 1d to 2d , -1 means the number of data is uncertain, the last 1 means color channel
x_image = tf.reshape(x,[-1,28,28,1])
#1st conv layer, 5x5 large ,1 channel, 32 masks
W_conv1 = weigth_varible([5,5,1,32])
b_conv1 = bias_varible([32])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1)+b_conv1)
h_pool1 = max_pool_2x2(h_conv1)
#2nd conv layer, still 5x5 large, after the 1st conv layer we get 32 channel, this time we use 64 masks
W_conv2 = weigth_varible([5,5,32,64])
b_conv2 = bias_varible([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2)+b_conv2)
h_pool2 = max_pool_2x2(h_conv2)
#1st fc_layer
#now the image is changed from 28x28x1 to 7x7x64, we will reshape it to 1d vector for the sake of full connection
#we decide to use 1024 hidden node
W_fc1 = weigth_varible([7*7*64,1024])
b_fc1 = bias_varible([1024])
h_pool2_flat = tf.reshape(h_pool2,[-1, 7*7*64])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat,W_fc1)+b_fc1)
#in case of overfitting, we add a dropout layer
keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
#softmax layer
W_fc2 = weigth_varible([1024,10])
b_fc2 = bias_varible([10])
y_conv = tf.nn.softmax(tf.matmul(h_fc1_drop,W_fc2)+b_fc2)
#loss-- cross entropy
cross_entropy = tf.reduce_mean(tf.reduce_sum(-y_*tf.log(y_conv), reduction_indices=[1]))
train_step = tf.train.AdagradOptimizer(1e-4).minimize(cross_entropy)
#test--
correct_pridiction = tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1))
accuracy = tf.reduce_mean(tf.cast(correct_pridiction, tf.float32))
#now ,start train
tf.global_variables_initializer().run()
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))
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}))
tensorflow写一个简单的CNN
最新推荐文章于 2023-07-25 09:25:47 发布