Python实现的递归神经网络简单示例
本文实例讲述了Python实现的递归神经网络。分享给大家供大家参考，具体如下：
#RecurrentNeuralNetworks
importcopy,numpyasnp
np.random.seed(0)
#computesigmoidnonlinearity
defsigmoid(x):
output=1/(1+np.exp(-x))
returnoutput
#convertoutputofsigmoidfunctiontoitsderivative
defsigmoid_output_to_derivative(output):
returnoutput*(1-output)
#trainingdatasetgeneration
int2binary={}
binary_dim=8
largest_number=pow(2,binary_dim)
binary=np.unpackbits(
np.array([range(largest_number)],dtype=np.uint8).T,axis=1)
foriinrange(largest_number):
int2binary[i]=binary[i]
#inputvariables
alpha=0.1
input_dim=2
hidden_dim=16
output_dim=1
#initializeneuralnetworkweights
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)
#traininglogic
forjinrange(10000):
#generateasimpleadditionproblem(a+b=c)
a_int=np.random.randint(largest_number/2)#intversion
a=int2binary[a_int]#binaryencoding
b_int=np.random.randint(largest_number/2)#intversion
b=int2binary[b_int]#binaryencoding
#trueanswer
c_int=a_int+b_int
c=int2binary[c_int]
#wherewe'llstoreourbestguess(binaryencoded)
d=np.zeros_like(c)
overallError=0
layer_2_deltas=list()
layer_1_values=list()
layer_1_values.append(np.zeros(hidden_dim))
#movingalongthepositionsinthebinaryencoding
forpositioninrange(binary_dim):
#generateinputandoutput
X=np.array([[a[binary_dim-position-1],b[binary_dim-position-1]]])
y=np.array([[c[binary_dim-position-1]]]).T
#hiddenlayer(input~+prev_hidden)
layer_1=sigmoid(np.dot(X,synapse_0)+np.dot(layer_1_values[-1],synapse_h))
#outputlayer(newbinaryrepresentation)
layer_2=sigmoid(np.dot(layer_1,synapse_1))
#didwemiss?...ifso,byhowmuch?
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])
#decodeestimatesowecanprint(itout)
d[binary_dim-position-1]=np.round(layer_2[0][0])
#storehiddenlayersowecanuseitinthenexttimestep
layer_1_values.append(copy.deepcopy(layer_1))
future_layer_1_delta=np.zeros(hidden_dim)
forpositioninrange(binary_dim):
X=np.array([[a[position],b[position]]])
layer_1=layer_1_values[-position-1]
prev_layer_1=layer_1_values[-position-2]
#erroratoutputlayer
layer_2_delta=layer_2_deltas[-position-1]
#errorathiddenlayer
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'supdateallourweightssowecantryagain
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(outprogress)
ifj%1000==0:
print("Error:"+str(overallError))
print("Pred:"+str(d))
print("True:"+str(c))
out=0
forindex,xinenumerate(reversed(d)):
out+=x*pow(2,index)
print(str(a_int)+"+"+str(b_int)+"="+str(out))
print("------------")
运行输出：
Error:[3.45638663] Pred:[00000001] True:[01000101] 9+60=1 ------------ Error:[3.63389116] Pred:[11111111] True:[00111111] 28+35=255 ------------ Error:[3.91366595] Pred:[01001000] True:[10100000] 116+44=72 ------------ Error:[3.72191702] Pred:[11011111] True:[01001101] 4+73=223 ------------ Error:[3.5852713] Pred:[00001000] True:[01010010] 71+11=8 ------------ Error:[2.53352328] Pred:[10100010] True:[11000010] 81+113=162 ------------ Error:[0.57691441] Pred:[01010001] True:[01010001] 81+0=81 ------------ Error:[1.42589952] Pred:[10000001] True:[10000001] 4+125=129 ------------ Error:[0.47477457] Pred:[00111000] True:[00111000] 39+17=56 ------------ Error:[0.21595037] Pred:[00001110] True:[00001110] 11+3=14 ------------
英文原文：https://iamtrask.github.io/2015/11/15/anyone-can-code-lstm/
更多关于Python相关内容感兴趣的读者可查看本站专题：《Python数学运算技巧总结》、《Python数据结构与算法教程》、《Python函数使用技巧总结》、《Python字符串操作技巧汇总》、《Python入门与进阶经典教程》及《Python文件与目录操作技巧汇总》
希望本文所述对大家Python程序设计有所帮助。