PLSPCAdfunction.rar_PLS python_PLS_python_PLS-CAD英文_Python pca代码

# -*- coding: utf-8 -*- from __future__ import division from numpy import* from xlwt import Workbook, Formula import xlrd import pandas as pd import copy import matplotlib.pyplot as plt def regularization(z): # 该函数将z正则化，并将其各维均值存入re【0】，各维标准差存入re【1】，输出标准化后矩阵和re矩阵 a1=len(z) ag=len(z[0]) pc=[] re=[[] for i in range(2)] i = 0 while i<a1: if linalg.norm(z[i])==0: pc.append(i) i+=1 u=0 z0=z[:] for i in pc: i=i-u z0.pop(i) u+=1 a1 = len(z0) z1 = [[0]*ag for i in range(a1)] i=0 while i < a1: j = 0 s1 = sum(z0[i]) s2 = len(z0[i]) s3 = 0 while j < ag: s3 = s3 + (z0[i][j] - s1 / s2) * (z0[i][j] - s1 / s2) j += 1 s3 = pow(s3 / (s2 - 1), 0.5) re[0].append(s1 / s2) re[1].append(s3) j = 0 while j < ag: z1[i][j] = (z0[i][j] - s1 / s2) / s3 j += 1 i += 1 return [z1, i, re] def yz(zx0,zy0,list,ix): # 该函数可以根据zx0，zy0根据PLS理论求出对应的t r u等向量 l0 = len(zx0[0]) l10 = len(zx0) l2 = len(zy0) l3=len(list) l5=len(list[0]) zx=[[] for i in range(l10)] zy = [[] for i in range(l2)] dss=[[] for i in range(l2)] diss = [[] for i in range(l2)] i=0 while i<l3: if i!=ix: for j in list[i]: k=0 while k<l10: zx[k].append(zx0[k][j]) k+=1 k=0 while k<l2: zy[k].append(zy0[k][j]) k+=1 i+=1 zx1=mat(zx) zy1 = mat(zy) ej0 = zx1 * zy1.T * zy1 * zx1.T ej2 = zy1 * zx1.T * zx1 * zy1.T ej1 = array(ej0) ej3 = array(ej2) #求出特征向量和特征值 nmtf, l01f = linalg.eig((ej1+ej1.T)/2) nmt1f, c0f = linalg.eig((ej3+ej3.T)/2) nmt = nmtf.real l01 = l01f.real nmt1 = nmt1f.real c0 = c0f.real a0 = len(nmt) a1 = len(nmt1) i = 0 m0 = nmt[0].real m1 = nmt1[0].real n0 = 0 n1 = 0 while i < a0: if m0 < nmt[i].real: m0 = nmt[i].real n0 = i i += 1 i = 0 while i < a1: if m1 < nmt1[i].real: m1 = nmt1[i].real n1 = i i += 1 i = 0 #选出最大特征值计算对应的t，r，u zx2 = array(zx1) l11=array(l01) l1 = l11[:,n0]/linalg.norm(l11[:,n0]) t = dot(zx2.T, l1) t1 = t.tolist() zy2 = array(zy1) c11=array(c0) c1 = c11[:,n1]/linalg.norm(c11[:,n1]) u = dot(zy2.T, c1) u1 = u.tolist() p = dot(zx2, t) / pow(linalg.norm(t), 2) r = dot(zy2, t) / pow(linalg.norm(t), 2) p1=p.tolist() r1=r.tolist() zcy3=[[] for i in range(l2)] i=0 while i<l5: j=0 while j<l2: zcy3[j].append(t1[i]*r1[j]) dss[j].append(pow(zcy3[j][i]-zy[j][i],2)) diss[j].append(0) j+=1 i+=1 li0=l1.tolist() return [dss,t1,u1,p1,r1,li0,diss] def yzh(zx0,zy0,list): # 该函数可以根据zx0，zy0根据PLS理论求出不同主成分数对应的误差和 l0 = len(zx0[0]) l10 = len(zx0) l2 = len(zy0) l3=len(list) l5=len(list[0]) nz=0 ssn=[[] for i in range(l3)] #根据交叉验证法计算不同主成分数误差和 while nz<l3: zx = [[] for i in range(l10)] zy = [[] for i in range(l2)] tz=[] uz=[] pz=[] rz=[] i=0 while i<l3: if i!=nz: for j in list[i]: k=0 while k<l10: zx[k].append(zx0[k][j]) k+=1 k=0 while k<l2: zy[k].append(zy0[k][j]) k+=1 i+=1 zcx = [[] for i in range(l10)] zcy = [[] for i in range(l2)] for i in list[nz]: j = 0 while j < l10: zcx[j].append(zx0[j][i]) j += 1 j = 0 while j < l2: zcy[j].append(zy0[j][i]) j += 1 h=0 zxi= mat(zx) zyi = mat(zy) zx1 = mat(zx) zy1 = mat(zy) while h<l10: ej0 = zx1 * zy1.T * zy1 * zx1.T ej2 = zy1 * zx1.T * zx1 * zy1.T ej1 = array(ej0) ej3 = array(ej2) #求出特征向量和特征值 nmtf, l01f = linalg.eig((ej1+ej1.T)/2) nmt1f, c0f = linalg.eig((ej3+ej3.T)/2) nmt = nmtf.real l01 = l01f.real nmt1 = nmt1f.real c0 = c0f.real a0 = len(nmt) a1 = len(nmt1) i = 0 m0 = nmt[0].real m1 = nmt1[0].real n0 = 0 n1 = 0 while i < a0: if m0 < nmt[i].real: m0 = nmt[i].real n0 = i i += 1 i = 0 while i < a1: if m1 < nmt1[i].real: m1 = nmt1[i].real n1 = i i += 1 i = 0 #选出最大特征值计算对应的t，r，u zx2 = array(zx1) l11=array(l01) l1 = l11[:,n0]/linalg.norm(l11[:,n0]) t = dot(zx2.T, l1) t1 = t.tolist() zy2 = array(zy1) c11=array(c0) c1 = c11[:,n1]/linalg.norm(c11[:,n1]) u = dot(zy2.T, c1) u1 = u.tolist() p = dot(zx2, t) / pow(linalg.norm(t), 2) r = dot(zy2, t) / pow(linalg.norm(t), 2) p1=p.tolist() r1=r.tolist() tz.append(t1) uz.append(u1) pz.append(p1) rz.append(r1) tz2=mat(tz) uz2=mat(uz) pz2=mat(pz) rz2=mat(rz) zcx1=mat(zcx) zcy1=mat(zcy) zcyc=zcx1.T*zxi*uz2.T*(tz2*zxi.T*zxi*uz2.T).I*tz2*zyi.T dyc=zcyc.T-zcy1 dyc0=dyc.tolist() dyc01=[[] for i in range(l2)] i=0 while i<l2: j=0 while j<len(dyc0[0]): dyc01[i].append(pow(dyc0[i][j],2)) j+=1 i+=1 ssi=sum(sum(dyc01)) zx1=(zxi.T-tz2.T*pz2).T zy1=(zyi.T-tz2.T*rz2).T ssn[nz].append(ssi) h+=1 nz+=1 return ssn def yz2(zx0,list,ix,hi): # 该函数可以根据zx0，zy0根据PCA理论求出对应的t l等向量 l0 = len(zx0[0]) l10 = len(zx0) l3=len(list) l5=len(list[0]) zx=[[] for i in range(l10)] i=0 while i<l3: if i!=ix: for j in list[i]: k=0 while k<l10: zx[k].append(zx0[k][j]) k+=1 i+=1 zx1=mat(zx) nn=eye(l10) nn1=nn/(l5-1) nn2=mat(nn1) ej0 =nn2* zx1* zx1.T ej1 = array(ej0) # 求出特征值和特征向量 nmtf, l01f = linalg.eig((ej1+ej1.T)/2) nmt0 = nmtf.real nmt=nmt0.tolist() l01 = l01f.real a0 = len(nmt) i = 0 m0 = nmt[0].real n0 = 0 n1 = 0 hj=0 hs=sum(nmt) if hi==1: nmt2=copy.deepcopy(nmt) i=0 hsi=0 while i<l5: nmax=max(nmt2) hsi=hsi+nmax nmt2.remove(nmax) a=hsi/hs if a>=0.85: