Tensorflow构建神经网络

先上代码

from __future__import print_function

import tensorflowas tf

import numpyas np

import matplotlib.pyplotas plt

#添加神经网络层的方法,如果有激活函数那么就是返回隐藏层,反之返回输出层.

def add_layer(inputs,in_size,out_size,activtion_function=None):

        Weights = tf.Variable(tf.random_normal([in_size,out_size]))  ## 权重

        biases=tf.Variable(tf.zeros([1,out_size])+0.1)  ## 偏置

        y=tf.matmul(inputs,Weights)+biases

        if activtion_functionis None:

            outputs = y

        else:

            outputs=activtion_function(y)

        return outputs

#构造神经网络的输入的数据

    x_data=np.linspace(-1,1,300)[:,np.newaxis]

    noise=np.random.normal(0,0.05,x_data.shape)

    y_data=np.square(x_data)-0.5+noise

#为神经网络输入定义 占位符 ,通过placeholder来定义。shape 参数可选，但是定义后tf可以自动捕捉# 数据维度不一致的错误

    xs=tf.placeholder(tf.float32,[None,1])

    ys=tf.placeholder(tf.float32,[None,1])

#添加隐藏层

    l1=add_layer(xs,1,10,activtion_function=tf.nn.relu)

#添加输出层

    prediction=add_layer(l1,10,1,activtion_function=None)

#损失函数

    loss=tf.reduce_mean(tf.reduce_sum(tf.square(ys-prediction),reduction_indices=[1]))

#优化策略,最小花误差函数

    train_step=tf.train.GradientDescentOptimizer(0.1).minimize(loss)

    init = tf.global_variables_initializer()

    sess=tf.Session()

    sess.run(init)

    fig=plt.figure()

    ax=fig.add_subplot(1,1,1)

    ax.scatter(x_data,y_data)

    plt.ion()

    plt.show()

    for i in range(1000):

        sess.run(train_step,feed_dict={xs:x_data,ys:y_data})

            if i %10 ==0:

        # to see the step improvement

        try:

            ax.lines.remove(lines[0])

        except Exception:

        pass

                prediction_value=sess.run(prediction,feed_dict={xs:x_data})

    #plot the prediction

        lines=ax.plot(x_data,prediction_value,'r-',lw=5)

        plt.pause(0.1)

sess.run(tf.global_variables_initializer()) 

这个方法在我们训练神经网络时都需要加上，官方的解释是初始化模型参数。参考：https://blog.csdn.net/u012436149/article/details/78291545。具体来说就是初始化Variable变量。

train_step=tf.train.GradientDescentOptimizer(0.1).minimize(loss)

梯度下降算法来最小化误差

添加隐藏层函数：

def add_layer(inputs,in_size,out_size,activtion_function=None):

        Weights = tf.Variable(tf.random_normal([in_size,out_size]))  ## 权重

        biases=tf.Variable(tf.zeros([1,out_size])+0.1)  ## 偏置

        y=tf.matmul(inputs,Weights)+biases

        if activtion_functionis None:

            outputs = y

        else:

            outputs=activtion_function(y)

        return outputs