tensorflow 非线性回归

import tensorflow as tfimport numpy as npimport matplotlib.pyplot as plt#线性回归#使用numpy生成200个随机点 变成200行1列x_data=np.linspace(-0.5,0.5,200)[:,np.newaxis]#生成一些干扰项noise=np.random.normal(0,0.2,x_data.shape)y_data=np.square(x_data)*noise#定义样本#行不确定，但是只有一列x=tf.placeholder(tf.float32,[None,1])y=tf.placeholder(tf.float32,[None,1])#构建简单的神经网络的中间层 1个输入神经元  10个中间神经元Weight_l1=tf.Variable(tf.random_normal([1,10]))baise_l1 = tf.Variable(tf.zeros([1,10]))#matmul矩阵的乘法Wx_plus_b_11=tf.matmul(x,Weight_l1)+baise_l1l1=tf.nn.tanh(Wx_plus_b_11)#定义输出层  中间层10个神经元 输出层1个神经元Weight_12=tf.Variable(tf.random_normal([10,1]))biase_12=tf.Variable(tf.zeros([1,1]))Wx_plus_b_12=tf.matmul(l1,Weight_12)+biase_12prediction=tf.nn.tanh(Wx_plus_b_12)#二次代价函数loss=tf.reduce_mean(tf.square(y-prediction))#使用梯度下降法train_step=tf.train.GradientDescentOptimizer(0.1).minimize(loss)with tf.Session() as sess:    sess.run(tf.global_variables_initializer())    for _ in range(2000):        sess.run(train_step,feed_dict={x:x_data,y:y_data})    #获得预测值    predict_value=sess.run(prediction,feed_dict={x:x_data})    #画图    plt.figure()    plt.scatter(x_data,y_data)    # r-代表是红色的实线   lw 代表线宽为5    plt.plot(x_data,predict_value,'r-',lw=5)    plt.show()