In [17]:

    

import tensorflow as tf
import numpy as np
tf.set_random_seed(55)


    

In [18]:

    

#构建网络
D_input = 2
D_label = 1
D_hidden = 2
lr=1e-4

x = tf.placeholder(tf.float32, [None, D_input], name="x")
t = tf.placeholder(tf.float32, [None, D_label], name="t")
    
W_h1 = tf.Variable(tf.truncated_normal([D_input, D_hidden], stddev=0.1), name="W_h")
b_h1 = tf.Variable(tf.constant(0.1, shape=[D_hidden]), name="b_h")
pre_act_h1 = tf.matmul(x, W_h1) + b_h1
act_h1 = tf.nn.relu(pre_act_h1, name='act_h')
    
W_o = tf.Variable(tf.truncated_normal([D_hidden, D_label], stddev=0.1), name="W_o")
b_o = tf.Variable(tf.constant(0.1, shape=[D_label]), name="b_o")
pre_act_o = tf.matmul(act_h1, W_o) + b_o
y = tf.nn.relu(pre_act_o, name='act_y')
    
loss=tf.reduce_mean((y-t)**2)
    
train_step = tf.train.AdamOptimizer(lr).minimize(loss)


    

In [19]:

    

#生成数据
X=[[0,0],[0,1],[1,0],[1,1]]
Y=[[0],[1],[1],[0]]
X=np.array(X).astype('int16')
Y=np.array(Y).astype('int16')


    

In [20]:

    

#加载
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())


    

In [21]:

    

#训练
for i in range(20000):
    sess.run(train_step,feed_dict={x:X,t:Y} )


    

In [22]:

    

#计算预测值
sess.run(y,feed_dict={x:X})


    

    
Out[22]:





array([[ 0.01419991],
       [ 0.97204113],
       [ 0.97204131],
       [ 0.04118853]], dtype=float32)

In [23]:

    

#查看隐藏层的输出
sess.run(act_h1,feed_dict={x:X})


    

    
Out[23]:





array([[ 0.96414495,  1.01730847],
       [ 0.        ,  0.50864619],
       [ 0.        ,  0.50864631],
       [ 0.        ,  0.        ]], dtype=float32)

In [24]:

    

#用完关闭session
sess.close()