tf-2. mnist 使用卷积方式来处理

源：http://www.tensorfly.cn/tfdoc/tutorials/mnist_pros.html

数据集：http://www.tensorfly.cn/tfdoc/tutorials/mnist_download.html

import tensorflow as tf

import input_data


minist = input_data.read_data_sets("MNIST_data/",one_hot=True)


        #y = tf.nn.softmax(wx+b)


x=tf.placeholder("float",[None,28*28])
x_=tf.reshape(x,[-1,28,28,1])


def conv(x):
#卷积1
    with tf.variable_scope('conv1'):
        w=tf.get_variable('weight',[5,5,1,32],initializer = tf.truncated_normal_initializer(stddev=0.1))
        b=tf.get_variable('bias',[32],initializer = tf.truncated_normal_initializer(stddev=0.1))
        conv1 = tf.nn.conv2d(x,w,strides=[1,1,1,1],padding="SAME")
        badd = tf.nn.bias_add(conv1,b)
        relu1 = tf.nn.relu(badd)
    with tf.name_scope("pool1"):
        pool1 = tf.nn.max_pool(relu1,[1,2,2,1],strides=[1,2,2,1],padding="SAME")


#卷积2
    with tf.variable_scope('conv2'):
        w=tf.get_variable('weight',[5,5,32,64],initializer = tf.truncated_normal_initializer(stddev=0.1)) #名不同
        b=tf.get_variable('bias',[64],initializer = tf.truncated_normal_initializer(stddev=0.1))
        conv2 = tf.nn.conv2d(pool1,w,strides=[1,1,1,1],padding="SAME")
        badd = tf.nn.bias_add(conv2,b)
        relu2 = tf.nn.relu(badd)
    with tf.name_scope("pool2"):
        pool2 = tf.nn.max_pool(relu2,[1,2,2,1],strides=[1,2,2,1],padding="SAME")


#展平
    t=tf.reshape(pool2,[-1,7*7*64])


    return t




#全连接层1
w=tf.Variable(tf.truncated_normal([7*7*64,1024],stddev=0.1))
b=tf.Variable(tf.truncated_normal(stddev=0.1,shape=[1024]))


x_flat = conv(x_)
relu1 = tf.nn.relu(tf.matmul(x_flat,w)+b)




#drop
keep_prob= tf.placeholder("float")
h_drop = tf.nn.dropout(relu1,keep_prob)


#全连接层2
w2=tf.Variable(tf.truncated_normal([1024,10],stddev = 0.1))
b2=tf.Variable(tf.truncated_normal(stddev=0.1,shape=[10]))


y=tf.nn.softmax(tf.matmul(h_drop,w2)+b2 ) #** w,b


y_=tf.placeholder("float",[None,10])


loss = -tf.reduce_sum(y_*tf.log(y))
train = tf.train.GradientDescentOptimizer(1e-4).minimize(loss) #学习率不同，导致错
correct_prediction = tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
accuracy=tf.reduce_mean(tf.cast(correct_prediction,"float"))




init = tf.global_variables_initializer()


with tf.Session() as sess:
    sess.run(init)
    for i in range(20000):
        batch_xs,batch_ys = minist.train.next_batch(50)
        sess.run(train,feed_dict={x:batch_xs,y_:batch_ys,keep_prob: 0.5})
        if i%100 == 0:
            train_accuracy = accuracy.eval(feed_dict={ x:batch_xs, y_: batch_ys, keep_prob: 1.0})
            print ("step %d, training accuracy %g"%(i, train_accuracy))


        
    correct_predict = tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
    accuracy=tf.reduce_mean(tf.cast(correct_predict,"float"))
    print(sess.run(accuracy,feed_dict={x:minist.test.images,y_:minist.test.labels}))

主要流程如下

图片输入->[卷积，池化]->[卷积，池化】->全连接层1024个隐藏结点->全连接层10个结点->softmax输出

卷积层：提取图片的特征，如下图，随着卷积层的进行，图像的特征，即下图的深度不断的加深

池化层：对卷积后的数据量进行缩小，如下图：池化层后，图像规格减半

drop：正则化，原于处理过拟合