基于RNN神经网络和BPTT算法实现的简单二进制计数器

import argparse
import time
import numpy as np
import copy


np.random.seed(0)


weightMat = [128, 64, 32, 16, 8, 4, 2, 1]


def sigmoid(x):
output = 1 / (1 + np.exp(-x))
return output




def sigmoid_derivative_value_byOutPut(output):
return output * (1 - output)




def binary2int(input):
if(len(input) != binary_dim):
print "input length is not " + str(binary_dim)
return -1

num = 0
for i in range(binary_dim):
num += input[i] * weightMat[i]

return num




int2binary = {}
binary_dim = 8


largest_number = pow(2,binary_dim)
binary = np.unpackbits(np.array([range(largest_number)], dtype = np.uint8).T, axis = 1)


for i in range(largest_number):
int2binary[i] = binary[i]




alpha = 0.1
input_dim = 2
hidden_dim = 16
output_dim = 1


MaxIterator = 20000


synapse_0 = 2 * np.random.random((input_dim, hidden_dim)) - 1
synapse_1 = 2 * np.random.random((hidden_dim, output_dim)) - 1
synapse_h = 2 * np.random.random((hidden_dim, hidden_dim)) - 1


synapse_0_update = np.zeros_like(synapse_0)
synapse_1_update = np.zeros_like(synapse_1)
synapse_h_update = np.zeros_like(synapse_h)


#train logic
for j in range(MaxIterator):


a_int = np.random.randint(largest_number / 2)
a = int2binary[a_int]


# print "a: " + str(a_int) 


b_int = np.random.randint(largest_number / 2)
b = int2binary[b_int]


# print "b: " + str(b_int)


c_int = a_int + b_int
c = int2binary[c_int]


d = np.zeros_like(c)


overallError = 0


layer_2_deltas = list()
layer_1_values = list()
layer_1_values.append(np.zeros(hidden_dim))


for position in range(binary_dim):


#generate input and output
x = np.array([[a[binary_dim - position - 1], b[binary_dim - position - 1]]])
y = np.array([[c[binary_dim - position - 1]]]).T




#hidden layer out value : layer_1
layer_1 = sigmoid(np.dot(x, synapse_0) + np.dot(layer_1_values[-1], synapse_h))


layer_2 = sigmoid(np.dot(layer_1, synapse_1))




layer_2_error = y - layer_2
layer_2_delta = layer_2_error * sigmoid_derivative_value_byOutPut(layer_2)

layer_2_deltas.append(layer_2_delta)


overallError += np.abs(layer_2_error[0])


d[binary_dim - position - 1] = np.round(layer_2[0][0])


layer_1_values.append(copy.deepcopy(layer_1))



future_layer_1_delta = np.zeros(hidden_dim)


for position in range(binary_dim):


x = np.array([[a[position], b[position]]])
layer_1 = layer_1_values[-position-1] 
pre_layer_1 = layer_1_values[-position-2]


layer_2_delta = layer_2_deltas[-position-1]  

layer_1_delta = (future_layer_1_delta.dot(synapse_h.T) + layer_2_delta.dot(synapse_1.T)) * sigmoid_derivative_value_byOutPut(layer_1)
synapse_1_update += np.atleast_2d(layer_1).T.dot(layer_2_delta)
synapse_0_update += x.T.dot(layer_1_delta)
synapse_h_update += np.atleast_2d(pre_layer_1).T.dot(layer_1_delta)


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 rnn train progress
if(j % 1000 == 0):
print "iterator : " + str(j)
print "Error: " + str(overallError)


print "Predict out: " + str(d) + "    " + str(binary2int(d)) 
print "True answer: " + str(c) + "    " + str(c_int)


out = 0
for index,x in enumerate(reversed(d)):
out += x * pow(2, index)


print str(a_int) + " + " + str(b_int) + " = " + str(out)
print "........................"




#if __name__ == '__main__':