• tensorflow函数介绍(3)


    tf.nn.softmax_cross_entropy_with_logits(logits,labels)      #其中logits为神经网络最后一层输出,labels为实际的标签,该函数返回经过softmax转换之后并与实际值相比较得到的交叉熵损失函数的值,该函数返回向量

    1、tf.nn.softmax_cross_entropy_with_logits的例子:

    import tensorflow as tf
    logits=tf.constant([[1.0,2.0,3.0],[1.0,2.0,3.0],[1.0,2.0,3.0]])  
    y=tf.nn.softmax(logits)      #计算给定输入的softmax值
    y_=tf.constant([[0.0,0.0,1.0],[0.0,0.0,1.0],[0.0,0.0,1.0]])  
    cross_entropy = -tf.reduce_sum(y_*tf.log(y))   #计算交叉熵损失函数的值,返回向量,并通过tf.reduce_sum来计算均值
    cross_entropy2=tf.reduce_sum(tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=y_))    #直接计算交叉熵损失函数值
    init=tf.global_variables_initializer()
    sess=tf.Session()
    sess.run(init)
    print(sess.run(y))
    print(sess.run(cross_entropy))    #输出结果和下面的一致
    print(sess.run(cross_entropy2))

     2、在tensorboard上显示运行图:

    import tensorflow as tf
    a = tf.constant(10,name="a")
    b = tf.constant(90,name="b")
    y = tf.Variable(a+b*2,name='y')
    init=tf.global_variables_initializer()
    with tf.Session() as sess:
        merged = tf.summary.merge_all()
        writer = tf.summary.FileWriter('/Users/jk/Desktop/2',sess.graph)     #自定义tensor到给定路径中
        sess.run(init)
        print(sess.run(y))

     通过在终端输入如下:

    cd 路径

    tensorboard --logdir=路径

    浏览器输入:http://localhost:6006/     得到tensorboard展示

    3、创建dropout层

    hidden1 = nn_layer(x, 784, 500, 'layer1')   #返回激活层的输出
    with tf.name_scope('dropout'): keep_prob = tf.placeholder(tf.float32) tf.summary.scalar('dropout_keep_probability', keep_prob) #添加到图 dropped = tf.nn.dropout(hidden1, keep_prob) #dropout

     4、参数的正则化 tf.contrib.layers.l2_regularizer(lambda)(w)

    返回给定参数的L1或L2正则化项的值

    w=tf.Variable(tf.random_normal([2,1],stddev=1,seed=1))
    x=tf.constant([1.,2.],shape=[1,2])
    y=tf.matmul(x,w)
    y_=tf.constant([1.,2.],shape=[1,2])
    lam=0.5
    loss=tf.reduce_mean(tf.square(y_-y))+tf.contrib.layers.l2_regularizer(lam)(w)
    tf.add_to_collection('loss',loss)
    z=tf.add_n(tf.get_collection('loss')) with tf.Session() as less: init
    =tf.global_variables_initializer() sess.run(init) print(sess.run(loss)) #输出总损失函数值
    print(sess.run(z)) #z是一般程序中较为常用的加入collection并给予输出的方式

      上述代码中的loss为损失函数,由两个部分组成,第一部分是均方误差损失函数,第二部分是正则化,用于防止模型过度模拟训练数据中的随机噪音。lam代表了正则化项的权重,也就是公式中的λ,其中w为需要计算正则化损失的参数。

      由于上述的定义方式会导致损失函数loss的定义很长,会导致可读性差,因而这时候就用到tensorflow提供的集合(collection),通过在一个计算图(graph)保存一组tensor。以如下代码为例:

    import tensorflow as tf
    def get_weight(shape,lam):        #通过collection逐个将每层的参数的正则化项进行收集
        var=tf.Variable(tf.random_normal(shape),dtype=tf.float32)
        tf.add_to_collection('losses',tf.contrib.layers.l2_regularizer(lam)(var))
        return var        
    x=tf.placeholder(tf.float32,shape=[None,2])   #设置placeholder作为每次的输入层
    y_=tf.placeholder(tf.float32,shape=[None,1])
    layer_dimension=[2,10,10,10,1]             #每一层的结点数
    n_layers=len(layer_dimension)              
    cur_layer=x                         #当前层
    in_dimension=layer_dimension[0]
    for i in range(1,n_layers):             
        out_dimension=layer_dimension[i]                        #提取当前输出层维度
        weight=get_weight([in_dimension,out_dimension],0.001)         #生成每层的权重,并汇总每层的正则化项
        bias=tf.Variable(tf.constant(0.1,shape=[out_dimension]))
        cur_layer=tf.nn.relu(tf.matmul(cur_layer,weight)+bias)        #计算每层经过激活函数relu后得到的输出值
        in_dimension=layer_dimension[i]                           #提取下一层的输入层维度
    mse_loss=tf.reduce_mean(tf.square(y_-cur_layer))
    tf.add_to_collection('losses',mse_loss)                  #将平方损失函数加入汇总损失函数
    loss=tf.add_n(tf.get_collection('losses'))               #提取集合losses里的所有部分都加起来
    with tf.Session() as sess:
        init=tf.global_variables_initializer()
        sess.run(init)
        print(sess.run(loss,feed_dict={x:[[1.,2.],[2.,3.]],y_:[[2.],[2.]]}))      #通过对占位符x和y_进行赋值,得到最终的损失值

     在如上运行完成后,再重新定义一次如下tf.placeholder()后,再进行一次sess.run(loss,feed_dict=....)却会报错,尚未找到原因。。。。(都是泪)

    x=tf.placeholder(tf.float32,shape=[None,2])   #设置placeholder作为每次的输入层
    y_=tf.placeholder(tf.float32,shape=[None,1])

     5、优化时设置记录全局步骤的单值,用于进一步的minimize

    global_step = tf.Variable(0, name='global_step', trainable=False)       #设置global_step变量不可训练
    train_op = optimizer.minimize(loss, global_step=global_step)

     6、计算准确率的常用套路(以mnist数字识别为例)

    import tensorflow as tf
    import numpy as np
    from tensorflow.examples.tutorials.mnist import input_data
    mnist=input_data.read_data_sets('data/',one_hot=True)
    trainimg=mnist.train.images
    trainlabel=mnist.train.labels
    testimg=mnist.test.images
    testlabel=mnist.test.labels
    x=tf.placeholder(tf.float32,[None,784])
    y=tf.placeholder(tf.float32,[None,10])
    w=tf.Variable(tf.zeros([784,10]))
    b=tf.Variable(tf.zeros([10]))
    pred=tf.nn.softmax(tf.matmul(x,w)+b)
    cost=tf.reduce_mean(-tf.reduce_sum(y*tf.log(pred),reduction_indices=[1]))     #计算交叉熵损失,reduce_sum为计算累加和,reduction_indices为计算的维度
    optm=tf.train.GradientDescentOptimizer(0.01).minimize(cost)
    corr=tf.equal(tf.argmax(pred,1),tf.argmax(y,1))     #返回预测值和实际值是否相同
    accr=tf.reduce_mean(tf.cast(corr,tf.float32))      #tf.cast()的作用是将corr的格式转为float32,然后通过tf.reduce_mean来计算平均准确率
    sess=tf.Session()       
    training_epochs=100
    training_epochs=100
    batch_size=100
    display_step=5
    init=tf.global_variables_initializer()
    sess.run(init)
    for epoch in range(training_epochs):
        avg_cost=0.
        num_batch=int(mnist.train.num_examples/batch_size)
        for i in range(num_batch):
            batch_xs,batch_ys=mnist.train.next_batch(batch_size)
            sess.run(optm,feed_dict={x:batch_xs,y:batch_ys})
            avg_cost+=sess.run(cost,feed_dict={x:batch_xs,y:batch_ys})/num_batch
        if epoch % display_step==0:
            train_acc=sess.run(accr,feed_dict={x:batch_xs,y:batch_ys})
            test_acc=sess.run(accr,feed_dict={x:mnist.test.images,y:mnist.test.labels})
            print('Epoch:%03d/%03d cost:%.9f Train accuracy: %.3f Test Accuracy: %.3f' %(epoch,training_epochs,avg_cost,train_acc,test_acc))  #字符串中%03d表示将对象epoch设定为长度为3的整型,%.3f表示取3位有效小数。%g表示保证6位有效数字前提下用小数方式,否则用科学计数法

     注:上面的这段代码运行的话需要去掉注释

    7、定义损失函数+优化器选择的常用套路

    cross_entropy=tf.nn.softmax_cross_entropy_with_logits(logits,ground_truth_input)
    cross_entropy_mean=tf.reduce_mean(cross_entropy)
    train_step=tf.train.GradientDescentOptimizer(learning_rate).minimize(corss_entropy_mean)

    8、每轮epoch均保存一次模型

    with tf.Session() as sess:
        init=tf.global_variables_initializer()
        sess.run(init)
        for i in range(training_steps):
            xs,ys=mnist.train.next_batch(batch_size)
            _,loss_value,step=sess.run([train_op,loss,global_step],feed_dict={x:xs,y_:ys})
            if i%1000==0:
                print('after %d training steps,loss on training batch is %g.' %(step,loss_value))                        
                saver.save(sess,os.path.join(model_save_path,model_name),global_step=global_step)    #可以让每个被保存模型的文件名末尾加上训练的轮数来进行记录
  • 相关阅读:
    Oracle主库存在Online Patch,备库该如何打上该补丁
    Oracle中如何构造一条在去年运行不报错今年运行报错的SQL语句
    Linux双网卡绑定启动网卡报错Error: Connection activation failed: Master connection not found or invalid
    Oracle备库GV$ARCHIVED_LOG.APPLIED的最新归档日志状态为"IN-MEMORY"(已经应用成功)对应主库的状态为"NO"
    Oracle关于ARCHIVELOG DELETION POLICY的配置解释以及RMAN-08137/RMAN-08591的原因探究
    Oracle Logminer的测试使用
    Oracle间隔分区(interval分区)的分区字段无法为NULL值
    Oracle绑定变量类型为timestamp导致V$SQL_BIND_CAPTURE不显示值
    即时性能分析工具 Pyroscope
    Go之Zap日志库集成Gin
  • 原文地址:https://www.cnblogs.com/xiaochouk/p/8453193.html
Copyright © 2020-2023  润新知