tensorflow的hello world

Posted 2021-02-11 z-bear

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


##定义网络结构
input_nodes  = 784
output_nodes = 10
layer1_nodes = 500

#定义超参数
#自动设置学习率
learning_rate_base=  0.8;
learning_decay = 0.99   ;
decay_step=100          ;

#滑动平均
moving_average__decay = 0.99
regularizer_rate  = 0.01;
train_step=30000
batch_size= 100


def inference(tensor1,weight1,bias1,weight2,bias2,average_class=None):
    if(average_class==None):
        layer1=tf.nn.relu(   tf.matmul(tensor1,weight1)+ bias1 )
        return tf.matmul( layer1,weight2 ) + bias2
    else:
        layer1 = tf.nn.relu(tf.matmul(tensor1, average_class.average(weight1)) + average_class.average(bias1))
        return tf.matmul(layer1, average_class.average(weight2) ) + average_class.average(bias2)

def get_weight(shape):
    weight=tf.Variable(tf.truncated_normal(shape=shape,stddev=0.1),tf.float32)
    tf.add_to_collection(‘losses‘, tf.contrib.layers.l2_regularizer(regularizer_rate)(weight))
    return weight

def get_bias(shape):
    return tf.Variable(tf.zeros(shape))

def train(mnist):
    #定义输入输出
    train_x=tf.placeholder(tf.float32,shape=[None,input_nodes],name=‘train_x‘)
    train_y=tf.placeholder(tf.float32,shape=[None,output_nodes],name=‘train_y‘ )

    weight1=get_weight( [input_nodes,layer1_nodes] )
    bias1   =get_bias([layer1_nodes])

    weight2=get_weight([layer1_nodes,output_nodes]);
    bias2  =get_bias([output_nodes])


    #定义学习率
    global_step = tf.Variable(0, trainable=False)
    learning_rate = tf.train.exponential_decay(learning_rate_base, global_step, decay_step, learning_decay,staircase=True)

    #定义损失、优化器
    results=inference(train_x,weight1,bias1,weight2,bias2,None)
    ce= tf.reduce_mean( tf.nn.sparse_softmax_cross_entropy_with_logits( logits=train_y,labels=tf.argmax( results) ) )
    loss=ce+tf.add_n( tf.get_collection(‘losses‘) )
    optimizer= tf.train.GradientDescentOptimizer(learning_rate).minimize(loss,global_step=global_step);

    #定义滑动平均
    ema = tf.train.ExponentialMovingAverage(moving_average__decay, global_step);
    maintain_average_op = ema.apply( tf.trainable_variables())
    with tf.control_dependencies([optimizer,maintain_average_op]):
        train_op=tf.no_op(name=‘train‘)

    #预测准确率
    average_y=inference(train_x,weight1,bias1,weight2,bias2,ema);
    correction_prediction = tf.equal(  tf.argmax( average_y,1 ) ,tf.argmax(train_y,1))
    accuracy = tf.reduce_mean(tf.cast(correction_prediction,tf.float32));

    with tf.Session() as sess:
        tf.global_variables_initializer().run()

        validate_feed={train_x:mnist.validation.images,train_y:mnist.validation.labels}
        test_feed    ={train_x:mnist.test.images,train_y:mnist.test.labels}


        #迭代训练
        for i in range(train_step):
            if(i%1000 == 0 ):
                validate_acc=sess.run(accuracy,feed_dict=validate_feed);
                print(‘After %d training steps,using aaverage model is %g ‘%(i,validate_acc))

            xt,yt=mnist.train.next_batch(batch_size);
            sess.run(train_op,feed_dict={ train_x :xt,train_y:yt});
        test_acc=sess.run(accuracy,feed_dict=test_feed)
        print(‘accuracy is %g‘%(test_acc));
def main():
    mnist= input_data.read_data_sets(‘./MNIST_data‘,one_hot=True)
    train(mnist);

if __name__ == ‘__main__‘:
    main()