用lstm实现7位数以内的加法

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
import numpy as np
import tensorflow as tf
from sklearn.preprocessing import OneHotEncoder
import random
import os
from tensorflow.contrib import rnn
data_num=np.array([0,1,2,3,4,5,6,7,8,9])
data_num=data_num.reshape([-1,1])
data_num=OneHotEncoder().fit_transform(data_num).todense()
batch_size=100
#print(data)
def get_data(batch_size=100):
    batch_in_data=[]
    batch_out_data=[]
    d1 = []
    for i in range(batch_size):
        a=random.randint(0,10000000)
        b=random.randint(0,10000000)
        c=[a,b]
        d=a+b
        
        d1.append(d)
        
        data=[]
        label=[]
        out_data=[]
        for j in c:
            for i in range(8):
                data.insert(i,j%10)
                j=j//10
        
        data=np.array(data)
        data=data.reshape([2,8]).T
        data=data.reshape([8,2])

        for i in range(8):
                out_data.insert(i,d%10)
                d=d//10      
        for x in out_data:
            label.append(data_num[x])
            
        #print(out_data)
        label=np.array(label)
        label=label.reshape([8,10])
        
        batch_out_data.append(label) 
        batch_in_data.append(data)
        
    batch_in_data=np.array(batch_in_data)
    batch_out_data=np.array(batch_out_data)
    batch_in_data=batch_in_data.astype("float32")
    batch_out_data=batch_out_data.astype("float32")
    batch_in_data=batch_in_data.reshape([-1,8,2])
    batch_out_data=batch_out_data.reshape([-1,8,10])
    
    return batch_in_data,batch_out_data,d1
#输入：
x=tf.placeholder(tf.float32,[None,8,2])
y=tf.placeholder(tf.float32,[None,8,10])
#y=tf.placeholder(tf.int64,[8])
'''
weights = {
    'out' : tf.Variable(tf.random_normal([8, 10])) #128*10
}

biases = {
    'out' : tf.Variable(tf.random_normal([10])) #1*10
}
'''
weights = {'in':tf.Variable(tf.random_normal([2, 8])),  
           'out':tf.Variable(tf.random_normal([8, 10]))}  
  
biases = {'in':tf.Variable(tf.constant(0.1, shape=[8, ])),  
           'out':tf.Variable(tf.constant(0.1, shape=[10, ]))}

def RNN(X, weights, biases):
    #print(X)
    #X=tf.transpose(X,[1,0,2])
    
    #X=tf.reshape(X,[-1,2])
    #X_in = tf.matmul(X, weights['in']) + biases['in']
    #print(X_in)
    #X_in = tf.reshape(X_in, [-1,8,16])
    #X_in=tf.reshape(X_in,[-1,8,8])
    
    X_in=tf.transpose(X,[1,0,2])
    
    #X_in = tf.split(x,8,0)
    print(X_in)
    # 使用基本的LSTM循环网络单元  
    #lstm_cell = tf.contrib.rnn.BasicLSTMCell(16, forget_bias=1.0, state_is_tuple=True)
    lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(8, forget_bias=1.0, state_is_tuple=True)
    #print(1)
    # 初始化为0，LSTM单元由两部分构成(c_state, h_state)  
    init_state = lstm_cell.zero_state(100, dtype=tf.float32)
    #print(2)
    # dynamic_rnn接收张量要么为(batch, steps, inputs)或者(steps, batch, inputs)作为X_in  
    outputs, final_state = tf.nn.dynamic_rnn(lstm_cell, X_in, initial_state=init_state, time_major=True)

    print("outputs:",outputs)
    
    outputs = tf.reshape(outputs,[-1,8])
    results = tf.matmul(outputs, weights["out"]) + biases["out"]
    #outputs = tf.reshape(outputs, [-1, 16])
    #results = tf.matmul(outputs, weights['out']) + biases['out']
    #print("results:",results)
    #results = tf.reshape(results, [-1, 8, 10])
    
    print("results:",results)
    results = tf.reshape(results,[8,-1,10])
    print(results)
    results = tf.transpose(results, [1, 0,2])
    #os.system("pause")
    return results

pred = RNN(x, weights, biases)

pre = tf.argmax(pred,2)
#pre = tf.argmax(pred,1)
#cost = tf.reduce_mean(tf.square(pred-y))
#print(pre)
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=pred, labels=y))
#cost = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=pred, labels=y))
#optimizer = tf.train.AdagradOptimizer(1).minimize(cost)

optimizer = tf.train.AdamOptimizer(0.01).minimize(cost)

correct_pred = tf.equal(tf.argmax(pred, 2), tf.argmax(y, 2))

accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

def train():
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for i in range(2000000):
            batch_xs,batch_ys,d=get_data(100)
            #print(batch_xs.shape,batch_ys.shape)

            sess.run(optimizer, feed_dict={x: batch_xs, y: batch_ys})
            
            #print(5)
            if i%100 ==0:
                cc = sess.run(accuracy, feed_dict={x: batch_xs, y: batch_ys})
                loss = sess.run(cost, feed_dict={x: batch_xs, y: batch_ys})
                print("cc:",cc," loss:",loss)

            if i % 1000==0:
                batch_xs,batch_ys,d=get_data(100)
                #print(batch_xs)
                #print(batch_ys)
                #os.system("pause")
                pree = sess.run(pre,feed_dict={x:batch_xs})
                #cc = sess.run(accuracy, feed_dict={x: batch_xs, y: batch_ys})
                #loss = sess.run(cost, feed_dict={x: batch_xs, y: batch_ys})
                for j in range(5):
                    print("true:",d[j])
                    print("pree:",pree[j][::-1][0:8])
                    print("________________")
                #os.system("pause")

        sess.close()

train()

1
以上就是实现的全过程，欢迎提问，有问必答，也欢迎指导