TensorFlow訓練神經網絡的4個步驟：

1、定義算法公式，即訓練神經網絡的forward時的計算

2、定義損失函數和選擇優化器來優化loss

3、訓練步驟

4、對模型進行準確率評測

附Multi-Layer Perceptron代碼：

 1 from tensorflow.examples.tutorials.mnist import input_data
 2 import tensorflow as tf
 3 
 4 mnist=input_data.read_data_sets("MNiST_data/",one_hot=True)
 5 sess=tf.InteractiveSession()
 
 6 
 7 in_units=784
 8 h1_units=300
 9 w1=tf.Variaable(tf.truncated_normal([in_units,h1_units],stddev=0.1))
10 b1=tf.Variable(tf.zeros([h1_units]))
11 w2=tf.Variable(tf.zeros([h1_units,10]))
12 b2=tf.Variable(tf.zeros([10]))
13 
14 x=tf.placeholder(tf.float32,[None,in_units])
15 keep_prob=tf.placeholder(tf.float32)
 
16 
17 hidden1=tf.nn.relu(tf.matmul(x,w1)+b1)
18 hidden1_drop=tf.nn.dropout(hidden1,keep_prob)
19 y=tf.nn.softmax(tf.matmul(hidden1_drop,w2)+b2)
20 
21 y_=tf.placeholder(tf.float32,[None,10])
22 cross_entropy=tf.reduce_mean(-tf.reduce_sum(y_*tf.log(y),reduction_indices=[1]))
23 train_step=tf.train.AdagradOptimizer(0.3).minimize(cross_entropy)
 
24 
25 tf.initialize_all_variables().run()
26 for i in range(3000):
27     batch_xs,batch_ys=mnist.train.next_batch(100)
28     train_step.run({x:batch_xs,y_:batch_ys,keep_prob:0.75})
29 
30 correct_prediction=tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
31 accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
32 print(accuracy.eval({x:mnist.test.images,y_:mnist.test.labels,keep_prob:1.0}))

TensorFlow-多層感知機(MLP)