小妹最近在写nsga2，目前正在将matlab的程式码，转换成python 但是一直遇到瓶颈，明明是写多目标，最後出来搞得跟单目标一样... 为何总是找到同一个解.... 尽自己最大的努力去转换程式语法， 网路上虽然有nsga2的程式码可以下载，但问题定义的方式有点落差 所以没办法直接拿现成的来用ＱＱ 程式码有点又臭又长，希望有人兄愿意尝试解救一下 是否能站内信交流呢～ 长得很丑的程式码就不放上来干扰视听了 下台鞠躬 p.s 另外请教板上众高手，小妹目前立马要对演算法熟悉以及熟悉python 是否有需要请高手来一对一教学？ 若版上高手认为可以针对小妹的问题来教学 也请欢迎站内信（如有违反版规，请板主通知，会删除这段） --

※ 发信站: 批踢踢实业坊(ptt.cc), 来自: 223.137.72.169 ※ 文章网址: https://webptt.com/cn.aspx?n=bbs/Python/M.1539704897.A.8CC.html ※ 编辑: majoyun (223.137.72.169), 10/17/2018 00:07:55 import numpy as np; import random; import math #问题定义 nvar=3; varsize=[1,nvar] varmin=-5; varmax=5 #变数上下界 ##cost适应函数 def costfun(x,nobj): n=np.size(x)#n为x的尺寸 z1=1-math.exp(-np.sum((x-1/math.sqrt(n))**2)) z2=1-math.exp(-np.sum((x+1/math.sqrt(n))**2)) z=[z1, z2] nobj=np.size(z) return z,nobj ##支配解函数 def Dom(x,y): x=np.array(x) y=np.array(y) b=((x<=y).all()) and ((x<y).any()) return b ##问题求解需要的结构建立 empty_sol_content={'position':list(), 'cost':list(), 'rank':list(), 'DominSet':int(),'DominCount':list(), 'CroDistance':list()} ##问题参数 problem1={'npop':50, 'pcoss':0.3, 'pmuta':0.9, 'mu':1,'maxit':100, 'sigma':0.1*(varmax-varmin)} npop=problem1['npop']; pcoss=problem1['pcoss']; pmuta=problem1['pmuta'] mu=problem1['mu']; sigma=problem1['sigma'];maxit=problem1['maxit'] problem2={'ncoss':2*round(pcoss*npop/2), 'nmuta':round(pmuta*npop)} ncoss=problem2['ncoss']; nmuta=problem2['nmuta'] ##产生初始解 pop=list()#初始解空集合 for i in range(0,npop): pop.append(empty_sol_content.copy()) pop[i]['position']=np.random.uniform(varmin,varmax,size=varsize) pop[i]['cost'],nobj=costfun(pop[i]['position'],nobj=list()) npop=np.size(pop) ##基因交叉函数 def crossover(x1,x2): y1=list();y2=list() alpha=np.random.uniform(0,1,size=np.size(x1)) y1=alpha*x1+(1-alpha)*x2 y2=alpha*x2+(1-alpha)*x1 return y1,y2 ##基因变异函数 def mutate(x,mu,sigma): sigma=int(sigma) nvar=np.size(x) x=x.reshape([1,3]) nmu=math.ceil(mu*nvar) j=np.random.choice(nvar,nmu) if np.size(sigma)>1: sigma=sigma[j] y=x y[0][j]=x[0][j]+np.dot(sigma,np.random.randn(np.size(j))) return y ####### po11=np.array([0.,0.,0.]) po22=np.array([0.,0.,0.]) for it in range(maxit): #popc1=np.tile(pop.copy(),int(ncoss/2)) popc1=pop[:int(ncoss/2)] #popc2=np.tile(empty_sol_content.copy(),int(ncoss/2)) popc2=pop[:int(ncoss/2)] popc1=list(popc1);popc2=list(popc2)#增加转list for k in range(int(ncoss/2)): i1=np.random.randint(0,npop) p1=pop[i1] i2=np.random.randint(0,npop) p2=pop[i2] po1,po2=crossover(p1['position'],p2['position']) po11=np.vstack((po11,po1)) po22=np.vstack((po22,po2)) po11=np.delete(po11,0,0) po22=np.delete(po22,0,0) po=np.vstack((po11,po22)) popc=np.hstack((popc1,popc2)) popc=list(popc) for k in range(len(popc)): popc[k]['position']=po[k] popc[k]['cost'],nobj=costfun(popc[k]['position'],nobj=list()) ####进入基因变异 #popm=np.tile(empty_sol_content.copy(),nmuta) popm=pop[:nmuta] popm=list(popm) for k in range(nmuta): i=np.random.randint(0,npop) p=pop[i] popm[k]['position']=mutate(p['position'],mu,sigma) popm[k]['cost'],nobj=costfun(popm[k]['position'],nobj=list()) pop=np.hstack((pop,popc)) pop=np.hstack((pop,popm)) pop=list(pop) #'DominedCount'被支配数、DominSet支配解 ##非支配排序 for i in range(len(pop)): pop[i]['DominSet']=[] pop[i]['DominCount']=0 F=list() Q=list() for i in range(len(pop)): for j in range(i+1,len(pop)): p=pop[i] q=pop[j] if Dom(p['cost'],q['cost']): p['DominSet']=np.hstack((p['DominSet'],j)) p['DominSet'].astype(np.int) q['DominCount']=q['DominCount']+1 if Dom(q['cost'],p['cost']): q['DominSet']=np.hstack((q['DominSet'],i)) q['DominSet'].astype(np.int) p['DominCount']=p['DominCount']+1 pop[i]=p pop[j]=q if pop[i]['DominCount']==0: F.append(i) pop[i]['rank']=0 Q.append(F)#把Ｆ变成一组放到Ｑ里面 F=Q F1=list() F1=F k=0 while True: Q=list() F=np.array(F) for i in F[0]: p=pop[i] for j in p['DominSet']: j=j.astype(np.int64) q=pop[j] q['DominCount']=q['DominCount']-1 if q['DominCount']==0: Q.append(j) q['rank']=k+1 pop[j]=q if Q==[]: break F1.append(Q) F=list() F.append(F1[k+1]) k=k+1 F=F1 ##拥挤距离 nf=len(F1) costs0=list() for k in range(nf): F=F1[k] costs=np.array([[0.],[0.]]) for f in range(np.size(F1[k])): costs0=pop[F[f]]['cost'][0]#把在Ｆ[k]中所有的cost都叫出来 costs1=pop[F[f]]['cost'][1] costs0=np.vstack((costs0,costs1))#转乘矩阵属性 costs=np.hstack((costs,costs0))#vstack合并在垂直方向上 costs=np.delete(costs,0,1)#把第一row删除（原本创建的[0,0]矩阵） nobj=len(np.mat(costs))#nobj为cost的row数 n=np.size(costs,1) d=np.zeros((n,nobj), dtype=np.float) cj=list() for j in range(nobj): cj=sorted(costs[j])#将costs由小至大排序 so=np.argsort(costs[j], axis=0)#排序後的costs下标情况 d[so[0],j]=np.inf if n-1>1: for i in range(1,n-1):#matlab i=2:n-1 d[so[i],j]=abs(cj[i+1]-cj[i-1])/abs(cj[0]-cj[-1]) d[so[-1],j]=np.inf d[so[-1],j]=np.inf for i in range(n): pop[F1[k][i]]['CroDistance']=sum(d[i,:])#F中的每一个都算好拥挤距离 ##拥挤距离计算完毕 ##sortingPopulation #先排rank rso=np.zeros(len(pop)) Rso=list() RSO=list() for i in range(len(pop)): rso[i]=pop[i]['rank']#rso里面是现在pop的rank状态 rso=rso.astype(np.int64) Rso=sorted(rso)#Rso把rso由小到大排序，接下来要知道排序的index RSO=np.argsort(rso,axis=0)#pop比须按照RSO顺序来排列rank才会由大到小 pop1=pop.copy()#pop1要把pop按照RSO重新排列 for i in range(len(pop)): for j in [RSO[i]]: pop1[i]=pop[j]#pop1已经将所有姐都按照rank小至大排列 ##再把每个rank中的拥挤距离由大到小排列 maxRso=max(Rso) Crod=list()#把crod排序之後放进来 CROD=list()#排序之後crod的index放进来 CDSO=np.zeros(len(pop)) pop2=pop.copy() for R in range(maxRso+1): crod=list() F=list() for i in range(len(pop1)): if pop1[i]['rank'] is R: crod.append(pop1[i]['CroDistance']) F=np.hstack((F,i)) crod=np.array(crod) Crod=crod[np.argsort(-crod)] CROD=np.argsort(-crod,axis=0) for f in range(len(F)): CDSO[int(F[f])]=F[CROD[f]] for i in range(len(pop)): pop2[i]=pop1[int(CDSO[i])]#pop2是用pop1的rank分组排列拥挤距离 ranks=[] for i in range(len(pop)): ranks=np.hstack((ranks,pop2[i]['rank'])) maxrank=max(ranks).astype(np.int64) F=list() F2=list() for r in range(maxrank+1): for i in range(len(pop2)): if pop2[i]['rank'] is r: F2.append(i) F.append(F2)#F是把同样rank的分成一组 F2=list() pop=pop2 ##F2 & pop2 是输出的参数 pop=pop[:npop] ########## ranks=[] for i in range(len(pop)): ranks=np.hstack((ranks,pop[i]['rank'])) maxrank=max(ranks).astype(np.int64) F=list() F2=list() for r in range(maxrank+1): for i in range(len(pop)): if pop2[i]['rank'] is r: F2.append(i) F.append(F2)#F是把同样rank的分成一组 F2=list() ##F2 & pop2 是输出的参数 ## store F1 F1=list() for i in F[0]: F1.append(pop[F[0][i]]) sf=[] for i in range(np.size(F1)): sf.append(F1[i]['cost']) sf=np.array(sf) #显示迭代资讯 print('Iteration:',it); print('Num of F1 Members:',np.size(F1)) import matplotlib.pyplot as plt x=sf[:,0] y=sf[:,1] x=list(x) y=list(y) plt.ion() plt.title('pcoss:0.7, pmuta:0.4, mu:0.2', fontsize=18) plt.scatter(x,y) #plt.xlim((0,1)) #plt.ylim((0,1)) plt.ioff() plt.pause(0.1) plt.show() ※ 编辑: majoyun (223.137.72.169), 10/17/2018 15:57:59