keras 使用RNN实现2进制加法

一.使用python实现RNN 2进制加法
这在很多文章中都出现了

#coding=utf-8
'''
Created on 2018年8月28日

'''

import copy, numpy as np                                       
np.random.seed(0)

# compute sigmoid nonlinearity                         #定义sigmoid函数
def sigmoid(x):
    output = 1/(1+np.exp(-x))
    return output

# convert output of sigmoid function to its derivative   #计算sigmoid函数的倒数
def sigmoid_output_to_derivative(output):
    return output*(1-output)

# training dataset generation                            
int2binary = {}                                    #用于将输入的整数转为计算机可运行的二进制数用
binary_dim = 8                                     #定义了二进制数的长度=8

largest_number = pow(2,binary_dim)                         #二进制数最大能取的数就=256喽
binary = np.unpackbits(
    np.array([range(largest_number)],dtype=np.uint8).T,axis=1)            
for i in range(largest_number):                                #将二进制数与十进制数做个一一对应关系
    int2binary[i] = binary[i]

# input variables
alpha = 0.1              #反向传播时参数w更新的速度
input_dim = 2            #输入数据的维度，程序是实现两个数相加的
hidden_dim = 16          #隐藏层神经元个数=16
output_dim = 1           #输出结果值是1维的


# initialize neural network weights                          #初始化神经网络的权重参数
synapse_0 = 2*np.random.random((input_dim,hidden_dim)) - 1   #输入至神经元的w0，维度为2X16，取值约束在[-1,1]间
synapse_1 = 2*np.random.random((hidden_dim,output_dim)) - 1  #神经元至输出层的权重w1,维度为16X1，取值约束在[-1,1]间
synapse_h = 2*np.random.random((hidden_dim,hidden_dim)) - 1  #神经元前一时刻状态至当前状态权重wh,维度为16X16，取值约束在[-1,1]间                                   

print(synapse_0.shape)
print(synapse_h.shape)
print(synapse_1.shape)

synapse_0_update = np.zeros_like(synapse_0)          #构造与w0相同维度的矩阵，并初始化为全0；
synapse_1_update = np.zeros_like(synapse_1)
synapse_h_update = np.zeros_like(synapse_h)

# training logic
for j in range(10000):          #模型迭代次数，可自行更改

    # generate a simple addition problem (a + b = c)
    a_int = np.random.randint(largest_number/2) # int version      #约束初始化的输入加数a的值不超过128
    a = int2binary[a_int] # binary encoding                        #将加数a的转为对应二进制数
    b_int = np.random.randint(largest_number/2) # int version
    b = int2binary[b_int] # binary encoding

    # true answer
    c_int = a_int + b_int    #真实和
    c = int2binary[c_int]    

    # where we'll store our best guess (binary encoded)
    d = np.zeros_like(c)                                        #用于存储预测的和

    overallError = 0                                           #打印显示误差

    layer_2_deltas = list()                                  #反向求导用
    layer_1_values = list()
    layer_1_values.append(np.zeros(hidden_dim))             #先对隐藏层前一时刻状态初始化为0

    # moving along the positions in the binary encoding
    for position in range(binary_dim):                         #前向传播；二进制求和，低位在右，高位在左

        # generate input and output
        X = np.array([[a[binary_dim - position - 1],b[binary_dim - position - 1]]]) #输入的a与b（二进制形式）
        y = np.array([[c[binary_dim - position - 1]]]).T                            #真实label值

        # hidden layer (input ~+ prev_hidden)
        layer_1 = sigmoid(np.dot(X,synapse_0) + np.dot(layer_1_values[-1],synapse_h))  # X*w0+RNN前一时刻状态值*wh

        # output layer (new binary representation)
        layer_2 = sigmoid(np.dot(layer_1,synapse_1))          #layer_1*w1

        # did we miss?... if so, by how much?
        layer_2_error = y - layer_2                                            #求误差
        layer_2_deltas.append((layer_2_error)*sigmoid_output_to_derivative(layer_2))  #代价函数
        overallError += np.abs(layer_2_error[0])                             #误差，打印显示用

        # decode estimate so we can print it out
        d[binary_dim - position - 1] = np.round(layer_2[0][0])                    #预测的和

        # store hidden layer so we can use it in the next timestep
        layer_1_values.append(copy.deepcopy(layer_1))                     #深拷贝，将RNN模块状态值存储，用于反向传播

    future_layer_1_delta = np.zeros(hidden_dim)

    for position in range(binary_dim):   #反向传播，计算从左到右，即二进制高位到低位

        X = np.array([[a[position],b[position]]])
        layer_1 = layer_1_values[-position-1]
        prev_layer_1 = layer_1_values[-position-2]

        # error at output layer
        layer_2_delta = layer_2_deltas[-position-1]
        # error at hidden layer
        layer_1_delta = (future_layer_1_delta.dot(synapse_h.T) + layer_2_delta.dot(synapse_1.T)) * sigmoid_output_to_derivative(layer_1)

        # let's update all our weights so we can try again
        synapse_1_update += np.atleast_2d(layer_1).T.dot(layer_2_delta)       #对w1进行更新
        synapse_h_update += np.atleast_2d(prev_layer_1).T.dot(layer_1_delta)  #对wh进行更新
        synapse_0_update += X.T.dot(layer_1_delta)                            #对w0进行更新

        future_layer_1_delta = layer_1_delta


    synapse_0 += synapse_0_update * alpha
    synapse_1 += synapse_1_update * alpha
    synapse_h += synapse_h_update * alpha    

    synapse_0_update *= 0
    synapse_1_update *= 0
    synapse_h_update *= 0

    # print out progress
    if(j % 1000 == 0):  #每1000次打印结果
        print ("Error:" + str(overallError))
        print ("Pred:" + str(d))
        print ("True:" + str(c))
        out = 0
        for index,x in enumerate(reversed(d)):
            out += x*pow(2,index)
        print (str(a_int) + " + " + str(b_int) + " = " + str(out))
        print ("------------")

二.使用keras实现上述RNN
上述一中的输入对应下面第3种输入数据

#coding=utf-8
'''
Created on 2018年8月27日

第一种输入数据:
77 + 11 = 88
3个整数的二进制分别为(低位->高危表示):
[0 1 0 0 1 1 0 1] [0 0 0 0 1 0 1 1] [0 1 0 1 1 0 0 0]
1-4列分别表示 00，10,01,11
X:
[[ 1.  0.  0.  0.]
 [ 0.  1.  0.  0.]
 [ 1.  0.  0.  0.]
 [ 1.  0.  0.  0.]
 [ 0.  0.  0.  1.]
 [ 0.  1.  0.  0.]
 [ 0.  0.  1.  0.]
 [ 0.  0.  0.  1.]]
Y:
[[0]
 [1]
 [0]
 [1]
 [1]
 [0]
 [0]
 [0]]

第二种输入数据:
60 + 72 = 132
3个整数的二进制分别为(低位->高危表示):
[0, 0, 1, 1, 1, 1, 0, 0] [0, 0, 0, 1, 0, 0, 1, 0] [0, 0, 1, 0, 0, 0, 0, 1]
1,2列表示60这个[0, 0, 1, 1, 1, 1, 0, 0]进行了one-hot编码
3,4列表示72这个[0, 0, 0, 1, 0, 0, 1, 0]进行了one-hot编码
X:
[[ 1.  0.  1.  0.]
 [ 1.  0.  1.  0.]
 [ 0.  1.  1.  0.]
 [ 0.  1.  0.  1.]
 [ 0.  1.  1.  0.]
 [ 0.  1.  1.  0.]
 [ 1.  0.  0.  1.]
 [ 1.  0.  1.  0.]]
Y:
[[0]
 [0]
 [1]
 [0]
 [0]
 [0]
 [0]
 [1]]

第三种输入数据:
3 + 23 = 26
3个整数的二进制分别为(低位->高危表示):
[1, 1, 0, 0, 0, 0, 0, 0] [1, 1, 1, 0, 1, 0, 0, 0] [0, 1, 0, 1, 1, 0, 0, 0]
1,2列分别表示3,23这两个数字的二进制
X:
[[1 1]
 [1 1]
 [0 1]
 [0 0]
 [0 1]
 [0 0]
 [0 0]
 [0 0]]
Y:
[[0]
 [1]
 [0]
 [1]
 [1]
 [0]
 [0]
 [0]]

第4种可以参考:https://blog.csdn.net/whai362/article/details/52523439
"789+123"这样的序列
'''
from keras.models import Sequential
from keras.layers import Activation, TimeDistributed, Dense,SimpleRNN
import numpy as np
from keras.utils import to_categorical
from keras.optimizers import SGD
from sklearn.model_selection import train_test_split

# training dataset generation                            
int2binary = {}                                    #用于将输入的整数转为计算机可运行的二进制数用
binary_dim = 8                                     #定义了二进制数的长度=8

largest_number = pow(2,binary_dim)                         #二进制数最大能取的数就=256喽
binary = np.unpackbits(
    np.array([range(largest_number)],dtype=np.uint8).T,axis=1)            
for i in range(largest_number):                                #将二进制数与十进制数做个一一对应关系
    int2binary[i] = binary[i]

x_train = []  
y_train = []  
for j in range(1):          #模型迭代次数，可自行更改
    # generate a simple addition problem (a + b = c)
    a_int = np.random.randint(largest_number/2) # int version      #约束初始化的输入加数a的值不超过128
    a = list(int2binary[a_int]) # binary encoding #将加数a的转为对应二进制数
    a.reverse()

    b_int = np.random.randint(largest_number/2) # int version
    b = list(int2binary[b_int]) # binary encoding
    b.reverse()

    c_int = a_int + b_int    #真实和
    c = list(int2binary[c_int])
    c.reverse()  

    #第2种输入数据，记得修改最后的keras model的输入维度
    #tempx = np.hstack((to_categorical(np.array([a]).T,num_classes=2),to_categorical(np.array([b]).T,num_classes=2)))

    #第3种输入数据，记得修改最后的keras model的输入维度
    #tempx = np.hstack((np.array([a]).T,np.array([b]).T))

    #第一种输入数据
    tempx = np.hstack((np.array([a]).T,np.array([b]).T))
    tempx_2 = []
    for i in range(8):
        if tempx[i][0] == 0 and tempx[i][1]==0:
            tempx_2.append(0)
        if tempx[i][0] == 1 and tempx[i][1]==0:
            tempx_2.append(1)
        if tempx[i][0] == 0 and tempx[i][1]==1:
            tempx_2.append(2)
        if tempx[i][0] == 1 and tempx[i][1]==1:
            tempx_2.append(3)
    tempx = to_categorical(np.array([tempx_2]).T,num_classes=4)

    tempy = np.array([c]).reshape((8,1))

    x_train.append(tempx)
    y_train.append(tempy)

x_train = np.array(x_train)
y_train = np.array(y_train)
x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=0.3, random_state=0)
print(x_train.shape)  
print(y_train.shape)




model = Sequential()
model.add(SimpleRNN(128, input_shape=(8, 4),return_sequences=True,activation="sigmoid"))
model.add(TimeDistributed(Dense(1)))
model.add(Activation('sigmoid'))
#经过测试最好不要用sgd,训练速度慢，建议使用adam
#sgd = SGD(lr=0.001, decay=0.0002, momentum=0.9, nesterov=True)
model.compile(loss='binary_crossentropy',
              optimizer='adam',
              metrics=['accuracy'])
model.summary() 

model.fit(x_train, y_train, 32, 1000,validation_data=(x_val, y_val))