Machine Learning 学习 之 RBF网络
来源：互联网 发布：校园网解绑网卡mac地址 编辑：程序博客网 时间：2024/05/16 19:43
RBF.py
#coding=utf8import mathimport randomimport numpy as npimport timeimport matplotlib.pyplot as pltrandom.seed(0)''' BP和RBF的思考： BP是有黑箱性质的东西,BP是基于插值和函数拟合的 RBF是逼近函数的一种核函数（径向基函数） RBF只具有感知范围内的探测性，不具有某些结构下（线性等）BP的特性 RBF是一种完全基于数据类似（KNN）方法的一个结构， 以径向基函数的插值代替构造的插值函数 故RBF的初始数据center非常重要，通常使用K-means 聚类实现'''def rand(a, b): return (b-a)*random.random() + a#不建议随机中心 由聚类确定class Layer: def __init__(self,typ,prev_num,num,Plist): self.typ=typ self.prev_num=prev_num self.num=num self.Nodes=[] self.Plist=Plist self.__createNodes() def __createNodes(self): for i in range(self.num): self.Nodes.append(Node(self.typ,i,self.prev_num,self.Plist))class Node: def __init__(self,typ,index,prev_num,Plist): self.typ=typ self.index=index self.prev_num=prev_num self.Plist=Plist self.__initNode() def __initNode(self): #如果是输入层 那么 一个神经元连接一个输入 w个数为1 if self.typ=='input': self.w=1 elif self.typ=='mm': self.tz=1 self.center=self.index+1 #print self.center #self.tz=rand(1,1) self.center=rand(0,3) elif self.typ=='output': self.w=[rand(-1,1) for i in range(self.prev_num)] #self.w=[0.30559951451034406, 1.8875762213796705] #self.w=[1,0] def __cal(self,data): inpu=0 if self.typ=='input': inpu=data[self.index] elif self.typ=='output': for i in range(self.prev_num): inpu+=self.w[i]*data[i] else: r=abs(self.center-data[0]) inpu=math.exp(-1.0*(r**2)/(self.tz**2)) #print self.center,data[0],r,inpu return inpu def calValue(self,data): self.inpu=self.__cal(data) self.out=self.inpu return self.outclass Net: def __init__(self): self.Layers=[] def setData(self,trainData): self.trainData=trainData def setNodes(self,NodesNum): prev_num=len(self.trainData[0][0]) layer=Layer('input',prev_num,prev_num,None) self.Layers.append(layer) self.NodesNum=NodesNum layer=Layer('mm',prev_num,self.NodesNum,None) self.Layers.append(layer) num=len(self.trainData[0][1]) layer=Layer('output',self.NodesNum,num,None) self.Layers.append(layer) def train(self,learn_rate,maxIter,epoch): #划分训练集 self.lr=learn_rate DATA=[self.trainData[i][0] for i in range(len(self.trainData))] RESULT=[self.trainData[i][1] for i in range(len(self.trainData))] for k in range(maxIter): #计算误差 循环 每个节点 每个训练样本 进行更新 error=[[0 for i in range(len(RESULT[0]))] for i in range(len(DATA))] for i in range(len(DATA)): dError=[0 for p in range(len(RESULT[0]))] prev_data=DATA[i] for layer in self.Layers: value=[] for node in layer.Nodes: #print prev_data,"SSSSSSSS" value.append(node.calValue(prev_data)) prev_data=value result=prev_data for j in range(len(RESULT[0])): #print len(RESULT),len(result),i,j dError[j]=RESULT[i][j]-result[j] error[i]=dError if k%epoch==0: print RESULT[i][j],result,dError for q in range(self.NodesNum): delta_center=0.0 delta_tz=0.0 delta_weight=0.0 sum1=0.0 sum2=0.0 sum3=0.0 center=self.Layers[1].Nodes[q].center tz=self.Layers[1].Nodes[q].tz for m in range(len(DATA)): X=DATA[m][0] sum1+=error[m][0]*math.exp(-1.0*(X-center)**2/(2*(tz**2)))*(X-center) sum2+=error[m][0]*math.exp(-1.0*(X-center)**2/(2*(tz**2)))*(X-center)**2 sum3+=error[m][0]*math.exp(-1.0*(X-center)**2/(2*(tz**2))) weight=self.Layers[2].Nodes[0].w[q] delta_center=self.lr*weight/(tz**2)*sum1 delta_delta=self.lr*weight/pow(tz,3)*sum2 delta_weight=self.lr*sum3 self.Layers[1].Nodes[q].center+=delta_center self.Layers[1].Nodes[q].tz+=delta_tz self.Layers[2].Nodes[0].w[q]+=delta_weight def test(self,testDATA): TestDATA=[testDATA[i][0] for i in range(len(testDATA))] TestRESULT=[testDATA[i][1] for i in range(len(testDATA))] predicts=[] Error=[0 for i in range(len(TestRESULT[0]))] for k in range(len(TestDATA)): result=self.predict(TestDATA[k]) predicts.append(result) for q in range(len(Error)): Error[q]=(Error[q]*k+abs(1.0*(TestRESULT[k][q]-result[q])))/(k+1) print TestRESULT[k][q],result[q],Error print Error self.__plotTest(TestRESULT,predicts) return Error def __plotTest(self,TestRESULT,predicts): Ax=[i for i in range(len(TestRESULT))] plt.plot(Ax,TestRESULT,label='true') plt.plot(Ax,predicts,label='predict') plt.legend(loc='upper left') plt.show() def predict(self,inpu): prev_data=inpu for layer in self.Layers: value=[] for node in layer.Nodes: value.append(node.calValue(prev_data)) prev_data=value result=prev_data return resulttest.py
#coding=utf8import timeimport mathfrom RBP import *trainData=[ [[1],[2]], [[1.1],[2.2]], [[1.3],[2.6]], [[1.4],[2.8]], [[1.6],[3.2]], [[1.7],[3.4]], [[1.8],[3.6]], [[1.9],[3.8]], [[2],[4]] ]def generate_data(n): """generate original data of u and y""" u = np.random.uniform(-5,5,n) y=[] #u=[] former_y_value = 0 for i in np.arange(0,n): #u.append(-5.0+10.0*i/n) #former_y_value=(u[i]+2*math.exp(-16*(u[i]**2)))/5 former_y_value=u[i]*math.sin(u[i])/5.0+u[i]**2/30-0.5 y.append(former_y_value) ''' next_y_value = 1.0*(29.0/40) * np.sin( (16.0 * u[i] + 8.0 * former_y_value) / (3 + 4 * (u[i] ** 2) + 4 * (former_y_value ** 2))) \ + (2.0/10) * u[i] + (2.0/10) * former_y_value #next_y_value=u[i] former_y_value = next_y_value ''' return u,yt0=time.clock()net1=Net()net1.setData(trainData)net1.setNodes(8)net1.train(0.008,20000,1000)print net1.predict([1.65])net1.test(trainData)print "消耗时间: ",time.clock()-t0,"s"阅读全文
0 0
- Machine Learning 学习 之 RBF网络
- CalTech machine learning, video 16 note(RBF)
- Machine Learning 之 Learning
- Machine Learning 学习 之 Qleaning 学习
- 系统学习Machine Learning之路漫漫
- Machine Learning 学习 之 Kmeans 笔记
- Machine Learning 学习之朴素贝叶斯
- Machine Learning学习之岭回归
- Machine Learning 学习之神经网络（1）
- Machine Learning 学习之决策树 ID3树
- Machine Learning 学习 之 C4.5
- Machine Learning 机器学习
- machine learning 学习网站
- 机器学习(Machine Learning)
- Machine learning学习(1)
- Machine Learning 学习笔记
- 机器学习 Machine Learning
- Machine Learning学习路线
- 洛谷10月月赛R2-T1-浮游大陆的68号岛
- 高阶函数（2）
- VMware虚拟机安装Linux操作系统
- LeetCode——503. Next Greater Element II
- 图像采集卡的种类和区别
- Machine Learning 学习 之 RBF网络
- 【服务器】owncloud的WebDAV
- CSS3 pointer-events:none应用举例及扩展
- 人工智能算法--KNN算法(C++实现)
- 字符串的选择性拷贝
- 在一个有序的旋转数组中，查找给定值
- Bribe the Prisoners(Easy)
- Machine Learning 学习 之 Qleaning 学习
- 基本运算符
