import tensorflow as tfimport numpy as npimport matplotlib.pyplot as pltplt.rcParams['font.sans-serif']=['FangSong'] # 用来正常显示中文标签plt.rcParams['axes.unicode_minus']=False# 用来正常显示负号plt.title('四阶多项式拟合')        #数据准备n_samples =100#数据集X = np.linspace(-3, 3, n_samples)#数据集标签，需要拟合的结果Y = np.sin(X) + np.random.uniform(0,1,n_samples)#静态图定义mg = tf.Graph()with mg.as_default():#占位xs = tf.placeholder(dtype=tf.float32,name='xs')ys = tf.placeholder(dtype=tf.float32,name='ys')#需要训练的权重变量w1 = tf.Variable(tf.random_normal([1]),name='w1')w2 = tf.Variable(tf.random_normal([1]),name='w2')w3 = tf.Variable(tf.random_normal([1]),name='w3')w4 = tf.Variable(tf.random_normal([1]),name='w4')b = tf.Variable(tf.zeros(1),name='b')#多项式公式my_pred = []my_pred.append(tf.multiply(w1,xs)) my_pred.append(tf.multiply(w2,tf.pow(xs,2)))my_pred.append(tf.multiply(w3,tf.pow(xs,3)))my_pred.append(tf.multiply(w4,tf.pow(xs,4)))y_pred = tf.add(tf.add_n(my_pred), b)#损失函数:使用样本方差loss = tf.reduce_sum(tf.pow(y_pred-ys, 2)) / (len(X)-1)#学习率，其实就是用多大的步进持续学习lr = 0.01#学习的轮数epochs = 1000#使用梯度下降法优化optimizer = tf.train.GradientDescentOptimizer(learning_rate=lr).minimize(loss)with tf.Session(graph=mg) as sess:sess.run(tf.global_variables_initializer())for epoch in range(epochs):total_loss = 0for x,y in zip(X,Y):_,tmp_loss = sess.run([optimizer,loss],feed_dict={xs:x,ys:y})total_loss += tmp_lossif epoch%10 == 0:print('epoch{:} the loss {:}'.format(epoch,total_loss))#获取系统优化结果mgw1,mgw2,mgw3,mgw4,mgb = sess.run([w1,w2,w3,w4,b])#session中的变量自动转换为全局变量print(mgw1,mgw2,mgw3,mgw4,mgb)#最终的拟合函数YY= X*mgw1 + X**2*mgw2 + X**3*mgw3 + X**4*mgw4 + mgbplt.plot(X, Y, 'bo', label='原始数据')plt.plot(X, YY, 'r', label='拟合数据')plt.legend()plt.show()