使用tensorflow实现最简单的线性回归算法

        #线性回归：用线性模型y=Wx+b拟合sin
        import numpy as np
        import matplotlib.pyplot as plt
        import tensorflow as tf
        
        #数据，标签
        x_data  = np.linspace(-2*np.pi,2*np.pi,300)
        noise = np.random.normal(-0.01,0.05,x_data.shape)
        y_label = np.sin(x_data) + noise
        plt.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('线性模型y=Wx+b拟合sin')
        plt.legend()
        plt.grid(True)
        plt.plot(x_data, y_label, 'b.', label='测试数据')
        
        #静态图定义
        mg = tf.Graph()
        with mg.as_default():
            #图输入
            X = tf.placeholder("float")
            Y = tf.placeholder("float")
            
            #训练权重w,b
            W = tf.Variable(np.random.randn(), name="weight")
            b = tf.Variable(np.random.randn(), name="bias")
            
            #线性模型y=Wx+b
            pred = tf.add(tf.multiply(X, W), b)
            
            #损失函数:使用样本方差
            cost = tf.reduce_sum(tf.pow(pred-Y, 2)) / (len(x_data)-1)
            
            #使用梯度下降法优化
            optimizer = tf.train.GradientDescentOptimizer(0.01).minimize(cost)
            
            #初始化图变量
            init = tf.group(tf.global_variables_initializer(),
                            tf.local_variables_initializer())
            
        with tf.Session(graph=mg) as sess:
            sess.run(init)
            for epoch in range(500):
                for (x, y) in zip(x_data, y_label):
                    sess.run(optimizer, feed_dict={X: x, Y: y})
                if (epoch+1) % 10 == 0:
                    c = sess.run(cost, feed_dict={X: x_data, Y:y_label})
                    print('epoch={:} cost={:0.6f} W={:0.6f} b={:0.6f}'.format(epoch+1,c,sess.run(W),sess.run(b)))
            training_cost = sess.run(cost, feed_dict={X: x_data, Y: y_label})
            print('训练结果 cost={:0.6f} W={:0.6f} b={:0.6f}'.format(training_cost,sess.run(W),sess.run(b)))
            plt.plot(x_data, sess.run(W) * x_data + sess.run(b),'r--' , label='拟合数据')
            plt.legend()
        plt.show()

 

---