Artificial Neural Network--Random Learning Perceptron

最近開始研讀類神經網路的相關文章
打算開發類神經網路訓練來解讀股價走勢
以下程式碼改自別人的範例
是我的蹲馬步練基本功的初步成果
最基本的神經網路--感知器



import random, math
import string

from sympy import *

def hardlims(inputM,weightM,bias):
     n=(weightM*inputM)[0].evalf()-bias
     if n<0:
     a = -1
     if n>=0:
     a = 1
     return a

def sum_array(a):
     acc = 0
     for i in a:
     acc += i
     return acc

class NN:
def __init__(self):
     self.w1 = [random.random() for x in range(10)]
     self.w2 = [random.random() for x in range(10)]
     self.w3 = [random.random() for x in range(10)]
     self.bias = 0

def learn(self,input,target):
     errorAbsSum = 1
     epsilon = 1e-6

     iterations = 0
     while abs(errorAbsSum) > epsilon:
     iterations +=1
     errorAbsSum = 0
     counter = 0
     tmp1 = (sum_array(self.w1) / len(self.w1))
     tmp2 = (sum_array(self.w2) / len(self.w2))
     tmp3 = (sum_array(self.w3) / len(self.w3))
     for i,x in enumerate(input):
     activation = hardlims(Matrix(3,1,x),Matrix(1,3,[tmp1,tmp2,tmp3]),self.bias)
     error = target[i] - activation
     learningRate = random.random() / 10
     tmp1 += (error * learningRate * x[0])
     tmp2 += (error * learningRate * x[1])
     tmp3 += (error * learningRate * x[2])
     errorAbsSum += abs(error)
     print 'Learning:w1=%f w2=%f w3=%f learningRate=%f error=%f activation=%f' % ( tmp1, tmp2,tmp3,learningRate,error,activation)
     del self.w1[0], self.w2[0],self.w3[0]
     self.w1.append(tmp1)
     self.w2.append(tmp2)
     self.w3.append(tmp3)
     print 'Iterations: %d \n w1=%f w2=%f w3=%f' % (iterations, tmp1, tmp2,tmp3)


def test(self,x):
     tmp1 = (sum_array(self.w1) / len(self.w1))
     tmp2 = (sum_array(self.w2) / len(self.w2))
     tmp3 = (sum_array(self.w3) / len(self.w3))
     activation = hardlims(Matrix(3,1,x),Matrix(1,3,[tmp1,tmp2,tmp3]),self.bias)
     print 'test(%s):result=%d w1=%f w2=%f w3=%f' % (x, activation, tmp1, tmp2,tmp3)


if __name__ == '__main__':
     input =[[-1,-1,-1],[1,1,1],[1,-1,-1]]
     target = [-1,1,1]

     nn=NN();
     nn.learn(input,target)
     nn.test([1,1,1])
     nn.test([1,1,-1])
     nn.test([1,-1,1])
     nn.test([1,-1,-1])
     nn.test([-1,1,1])
     nn.test([-1,1,-1])
     nn.test([-1,-1,1])
     nn.test([-1,-1,-1])


驗收訓練成果

test([1, 1, 1]):result=1 w1=0.663924 w2=0.276574 w3=0.333518
test([1, 1, -1]):result=1 w1=0.663924 w2=0.276574 w3=0.333518
test([1, -1, 1]):result=1 w1=0.663924 w2=0.276574 w3=0.333518
test([1, -1, -1]):result=1 w1=0.663924 w2=0.276574 w3=0.333518
test([-1, 1, 1]):result=-1 w1=0.663924 w2=0.276574 w3=0.333518
test([-1, 1, -1]):result=-1 w1=0.663924 w2=0.276574 w3=0.333518
test([-1, -1, 1]):result=-1 w1=0.663924 w2=0.276574 w3=0.333518
test([-1, -1, -1]):result=-1 w1=0.663924 w2=0.276574 w3=0.333518


縮排有點亂
有誰能指點一下?