tensorflow实现mnist手写数字识别

非卷积实现

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data


data_path = 'F:\CNN\data\mnist'

mnist_data = input_data.read_data_sets(data_path,one_hot=True) #offline dataset

x_data = tf.placeholder("float32", [None, 784]) # None means we can import any number of images


weight =  tf.Variable(tf.ones([784,10]))

bias = tf.Variable(tf.ones([10]))

Y_model = tf.nn.softmax(tf.matmul(x_data ,weight) + bias)

#Y_model = tf.nn.sigmoid(tf.matmul(x_data ,weight) + bias)


'''
weight1 =  tf.Variable(tf.ones([784,256]))

bias1 = tf.Variable(tf.ones([256]))

Y_model1 = tf.nn.softmax(tf.matmul(x_data ,weight1) + bias1)

weight1 =  tf.Variable(tf.ones([256,10]))

bias1 = tf.Variable(tf.ones([10]))

Y_model = tf.nn.softmax(tf.matmul(Y_model1 ,weight1) + bias1)
'''



y_data = tf.placeholder("float32", [None, 10])

loss = tf.reduce_sum(tf.pow((y_data - Y_model), 2 ))#92%-93%

#loss = tf.reduce_sum(tf.square(y_data - Y_model)) #90%-91%

optimizer = tf.train.GradientDescentOptimizer(0.01)
train = optimizer.minimize(loss)

init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init) # reset values to wrong
for i in range(100000):
    batch_xs, batch_ys = mnist_data.train.next_batch(50)
    sess.run(train, feed_dict = {x_data: batch_xs, y_data: batch_ys})
    if i%50==0:
        correct_predict = tf.equal(tf.arg_max(Y_model,1),tf.argmax(y_data,1))
        accurate = tf.reduce_mean(tf.cast(correct_predict,"float"))
        print(sess.run(accurate,feed_dict={x_data:mnist_data.test.images,y_data:mnist_data.test.labels}))

卷积实现

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data


data_path = 'F:\CNN\data\mnist'

mnist_data = input_data.read_data_sets(data_path,one_hot=True) #offline dataset

x_data = tf.placeholder("float32", [None, 784]) # None means we can import any number of images

x_image = tf.reshape(x_data, [-1,28,28,1])

w_conv = tf.Variable(tf.ones([5,5,1,32]))  #weight

b_conv = tf.Variable(tf.ones([32]))        #bias

h_conv = tf.nn.relu(tf.nn.conv2d(x_image , w_conv,strides=[1,1,1,1],padding='SAME')+ b_conv)

h_pool = tf.nn.max_pool(h_conv,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')

w_fc = tf.Variable(tf.ones([14*14*32,1024]))

b_fc = tf.Variable(tf.ones([1024]))

h_pool_flat = tf.reshape(h_pool,[-1,14*14*32])

h_fc = tf.nn.relu(tf.matmul(h_pool_flat,w_fc) +b_fc)

W_fc = w_fc = tf.Variable(tf.ones([1024,10]))

B_fc = tf.Variable(tf.ones([10]))

Y_model = tf.nn.softmax(tf.matmul(h_fc,W_fc) +B_fc)

y_data = tf.placeholder("float32",[None,10])

loss =  -tf.reduce_sum(y_data * tf.log(Y_model))

train_step = tf.train.GradientDescentOptimizer(0.01).minimize(loss)

init = tf.initialize_all_variables()

sess = tf.Session()

sess.run(init)

for i in range(1000):
    batch_xs,batch_ys =mnist_data.train.next_batch(5)
    sess.run(train_step,feed_dict={x_data:batch_xs,y_data:batch_ys})
    if i%50==0:
        correct_prediction = tf.equal(tf.argmax(Y_model,1),tf.argmax(y_data,1))
        accuracy = tf.reduce_mean(tf.cast(correct_prediction,"float"))
        print(sess.run(accuracy,feed_dict={x_data:mnist_data.test.images,y_data:mnist_data.test.labels}))