RNN的小demo

详解请看地址：http://iamtrask.github.io/2015/11/15/anyone-can-code-lstm/

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

# compute sigmoid nonlinearity

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

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

int2binary = {}
binary_dim = 8
largest_number = pow(2,binary_dim) # 2的8次方 256

intnum=np.array([range(largest_number)])
#print(intnum)
binary = np.unpackbits(np.array([range(largest_number)],dtype=np.uint8).T,axis=1) #256个数字，变成2进制后每个十进制变成了8位  256*8=2048
#print(binary,binary.size)
for i in range(largest_number):
 int2binary[i] = binary[i]

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

synapse_0 = 2*np.random.random((input_dim,hidden_dim)) - 1 #服从正态分布  2行16列
synapse_1 = 2*np.random.random((hidden_dim,output_dim)) - 1 #16行1列
synapse_h = 2*np.random.random((hidden_dim,hidden_dim)) - 1 #16行 16列

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

for j in range(10):
    a_int = np.random.randint(largest_number/2) # int version   左闭右开整数  生成一个随机数
    a = int2binary[a_int] # binary encoding
    b_int = np.random.randint(largest_number/2) # int version  左闭右开整数  生成一个随机数
    b = int2binary[b_int] # binary encoding
    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() #[0,0,0,0,0,0,0,0,0,0,0,0,0]
    layer_1_values.append(np.zeros(hidden_dim))

    for position in range(binary_dim):
        X = np.array([[a[binary_dim - position - 1],b[binary_dim - position - 1]]])
        y = np.array([[c[binary_dim - position - 1]]]).T
        layer_1 = sigmoid(np.dot(X,synapse_0) + np.dot(layer_1_values[-1],synapse_h)) # layer_1_values[-1]  取最后的 [2,16].dot[16,16]
        layer_2 = sigmoid(np.dot(layer_1,synapse_1))
        layer_2_error = y - layer_2
        layer_2_deltas.append((layer_2_error)*sigmoid_output_to_derivative(layer_2))
        overallError += np.abs(layer_2_error[0])
        d[binary_dim - position - 1] = np.round(layer_2[0][0]) #np.round 四舍五入
        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]
            prev_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_output_to_derivative(layer_1)
            synapse_1_update += np.atleast_2d(layer_1).T.dot(layer_2_delta)
            synapse_h_update += np.atleast_2d(prev_layer_1).T.dot(layer_1_delta)
            synapse_0_update += X.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 progress
    if(j % 1000 == 0):  #每10次输出一次
        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 ("------------")