tensorflow实现回归

直线拟合：y=w*x+b

"""
回归:直线拟合
"""
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
x_data=np.random.rand(100)
y_data=x_data*5+1W=tf.Variable(0.)
b=tf.Variable(0.)
y_pred=W*x_data+bx=tf.placeholder(shape=None,dtype=tf.float32)loss=tf.reduce_mean(tf.square(y_data-y_pred))
optimizer=tf.train.GradientDescentOptimizer(0.3).minimize(loss)
with tf.Session() as sess:sess.run(tf.global_variables_initializer())for i in range(100):sess.run(optimizer)if i %10==0:W1,b1=sess.run([W,b])print('step={},W={},b={}'.format(i,W1,b1))prediction=sess.run(y_pred,feed_dict={x:x_data})plt.scatter(x_data,y_data)plt.plot(x_data,prediction)plt.show()

二，二次拟合 :y=w*x*x+b ,一层hidden layer,10个节点，一个output一个节点

"""
回归：二次拟合
"""
import tensorflow as tf
import  numpy as np
import matplotlib.pyplot as plt
#生成两百个随机点
x_data=np.linspace(-1,1,200).reshape([-1,1]).astype(np.float32)
noise=np.random.normal(loc=0.,scale=0.02,size=x_data.shape)
y_data=np.square(x_data)+noisex=tf.placeholder(shape=[None,1],dtype=tf.float32)
y=tf.placeholder(shape=[None,1],dtype=tf.float32)
#hidden_layer
W1=tf.Variable(tf.random_normal(shape=[1,10],stddev=tf.sqrt(2.)),dtype=tf.float32)
b1=tf.Variable(tf.zeros(shape=[1,10]),dtype=tf.float32)#output_layer
W2=tf.Variable(tf.random_normal(shape=[10,1],stddev=tf.sqrt(2./10)),dtype=tf.float32)
b2=tf.Variable(tf.zeros(shape=[1,1]),dtype=tf.float32)Z1=tf.matmul(x_data,W1)+b1
A1=tf.nn.relu(Z1)
y_pred=tf.matmul(A1,W2)+b2
# y_pred=tf.nn.relu(Z2)loss=tf.reduce_mean(tf.reduce_sum(tf.square(y-y_pred),axis=1))
optimizer=tf.train.GradientDescentOptimizer(0.2).minimize(loss)with tf.Session() as sess:sess.run(tf.global_variables_initializer())costs=[]for i in range(2000):cost,_=sess.run([loss,optimizer],feed_dict={x:x_data,y:y_data})if i%100==0:costs.append(cost)prediction = sess.run(y_pred, feed_dict={x: x_data})plt.scatter(x_data, y_data)plt.plot(x_data,prediction)plt.show()plt.plot(costs)plt.show()

打印结果：