R 의사결정나무, 인공신경망

###의사결정나무 auto=read.csv('autoparts.csv') auto1=auto[auto$prod_no=='90784-76001',c(2:11)] auto2=auto1[auto1$c_thickness<1000,] auto2$y_falty=ifelse((auto2$c_thickness<20)|(auto2$c_thickness>32),1,0) #불량(1)에 관심==positive t_index=sample(1:nrow(auto2),size=nrow(auto1)*0.7) train=auto2[t_index,] test=auto2[-t_index,] nrow(train);nrow(test) head(train) install.packages('tree') library(tree) m=tree(factor(y_falty)~fix_time+a_speed+b_speed+separation+s_separation+          rate_terms+mpa+load_time+highpressure_time,data=train) plot(m) text(m) #가지치기 prune.m=prune.tree(m,method='misclass') #misclass 잘못된분류 plot(prune.m) prune.m9=prune.tree(m,best=9) #가짓수 9개 plot(prune.m9) text(prune.m9) prune.m3=prune.tree(m,best=3) plot(prune.m3) text(prune.m3) #분류모형 정확도 검증_크로스테이블 만들기 yhat_test=predict(m,test,type='class') #분류이기 때문에 타입이 class table=table(real=test$y_falty,predict=yhat_test);table (table[1,1]+table[2,2])/sum(table) #정분류율 yhat_test=predict(prune.m3,test,type='class') #분류이기 때문에 타입이 class table=table(real=test$y_falty,predict=yhat_test);table (table[1,1]+table[2,2])/sum(table) yhat_test=predict(prune.m9,test,type='class') #분류이기 때문에 타입이 class table=table(real=test$y_falty,predict=yhat_test);table (table[1,1]+table[2,2])/sum(table) #ROC, AUC - ROC커브 그린 후 AUC값 꼭 확인하기 library(Epi) ROC(test=yhat_test,stat=test$y_falty,plot='ROC',AUC=T,main='TREE') #예측 new.data=data.frame(fix_time=87,a_speed=0.609,b_speed=1.715,separation=242.7,                     s_separation=657.5,rate_terms=95,mpa=78,load_time=18.1,highpressure_time=82) predict(m,newdata = new.data,type='class') new.data=data.frame(fix_time=c(87,85.6),a_speed=c(0.609,0.472),b_speed=c(1.715,1.685),                     separation=c(242.7,243.4),s_separation=c(657.5,657.9),rate_terms=c(97,95),                     mpa=c(78,28.8),load_time=c(18.1,18.2),highpressure_time=c(82,60)) predict(m,newdata = new.data,type='class') new.data=data.frame(fix_time=test$fix_time,a_speed=test$a_speed,b_speed=test$b_speed,                     separation=test$separation,s_separation=test$s_separation,                     rate_terms=test$rate_terms,mpa=test$mpa,load_time=test$load_time,                     highpressure_time=test$highpressure_time) predict(m,newdata=new.data,tyle='class') #종속변수 범주가 다항인 경우(2만 정상, 1과3은 불량) auto2$gclass=as.factor(ifelse(auto2$c_thickness<20,1,ifelse(auto2$c_thickness<32,2,3))) t_index=sample(1:nrow(auto2),size=1:nrow(auto2)*0.7) train=auto2[t_index,] test=auto2[-t_index,] ##############################안되는 부분 m<-tree(gclass~fix_time+a_speed+b_speed+separation+s_separation+          rate_terms+mpa+load_time+highpressure_time,data=train) yhat_test=predict(m,test,type='class') table=table(real=test$g_class,predict=yhat_test (table[1,1]+table[2,2])/sum(table) m=tree(c_thickness~fix_time+a_speed+b_speed+separation+s_separation+          rate_terms+mpa+load_time+highpressure_time,data=train) plot(m) text(m) ######################################################## #mse: 예측값-실제값==오차 제곱해서 평균 mse=mean((yhat_test-test$c_thickness)^2) #------------------------------- ###K-근접 이웃 분류(K-NN) auto=read.csv('autoparts.csv') auto1=auto[auto$prod_no=='90784-76001',c(2:11)] auto2=auto1[auto1$c_thickness<1000,] auto2$y_faulty=ifelse((auto2$c_thickness<20)|(auto2$c_thickness>32),1,0) t_index=sample(1:nrow(auto2),size=nrow(auto2)*0.7) train=auto2[t_index,] test=auto2[-t_index,] xmat.train=as.matrix(train[1:9]) #1열부터 9열까지 매트릭스화 처리 y_faulty.train=train$y_faulty xmat.test=as.matrix(test[1:9]) head(xmat.test) install.packages('class') library(class) yhat_test=knn(xmat.train,xmat.test,as.factor(y_faulty.train),k=3) yhat_test #테스트 데이터 셋의 종속변수 모음 table=table(real=test$y_faulty,predict=yhat_test) (table[1,1]+table[2,2])/sum(table)  ##knn 예측함수 ##최적값 k찾기 library(e1071) tune.out=tune.knn(x=xmat.train,y=as.factor(y_faulty.train),k=3) tune.out plot(tune.out) yhat_test=knn(xmat.train,xmat.test,y_faulty.train,k=5) table=table(real=test$y_faulty,predict=yhat_test);table (table[1,1]+table[2,2])/sum(table) library(Epi) ROC(test=yhat_test,stat=test$y_faulty,plot="ROC",AUC=T,main='KNN') new.data=data.frame(fix_time=87,a_spped=0.609,b_speed=1.715,separation=242.7,                     s_separation=657.5,rate_terms=95,mpa=18.1,highpressure_tiem=82) knn(xmat.train,new.data,y_faulty.train,k=5) auto=read.csv('autoparts.csv') auto1=auto[auto$prod_no=='90784-76001',c(2:11)] auto2=auto1[auto1$c_thickness<1000,] auto2$g_class=as.factor(ifelse(auto2$c_thickness<20),1,ifelse(auto2$c_thickness<32,2,3))) auto=read.csv('autoparts.csv') auto1=auto[auto$prod_no=='90784-76001',c(2:11)] auto2=auto1[auto1$c_thickness<1000,] auto2$g_class=as.factor(ifelse(auto2$c_thickness<20,1,ifelse(auto2$c_thickness<32,2,3))) t_index=sample(1:nrow(auto2),size=nrow(auto2)*0.7) train=auto2[t_index,] test=auto2[-t_index,] #예측값 생성 xmat.train=as.matrix(train[1:9]) g_class.train=train$g_class xmat.test=as.matrix(test[1:9]) library(class) yhat_test=knn(xmat.train,xmat.test,g_class.train,k=3) table=table(real=test$g_class,predict=yhat_test) (table[1,1]+table[2,2]+table[3,3])/sum(table) #knn으로 회귀분석하기 위해서는 매트릭스화 필요 ##KNN회귀분석 package'FNN'::knn.reg 이용 x.mat.train=as.matrix(train[1:9]) c_thickness.train=train$c_thickness xmat.test=as.matrix(test[1:9]) install.packages('FNN') library(FNN) yhat_test=knn.reg(xmat.train,xmat.test,c_thickness.train,k=3) mse=mean((yhat_test$pred-test$c_thickness)^2);mse ###인공신경망  auto=read.csv('autoparts.csv') auto1=auto[auto$prod_no=='90784-76001',c(2:11)] auto2=auto1[auto1$c_thickness<1000,] auto2$g_class=as.factor(ifelse(auto2$c_thickness<20,1,ifelse(auto2$c_thickness<32,2,3))) t_index=sample(1:nrow(auto2),size=nrow(auto2)*0.7) train=auto2[t_index,] test=auto2[-t_index,] install.packages('nnet') library(nnet) m=nnet(g_class ~ fix_time+a_speed+b_speed+separation+s_separation+rate_terms+mpa+          load_time+highpressure_time,data=train,size=10) #성능평가_크로스테이블_오차표 yhat_test=predict(m,test,type='class') #분류이기 때문에 타입이 class인 것. table=table(real=test$g_class,predict=yhat_test) table (table[1,1]+table[2,2]+table[3,3])/sum(table) new.data=data.frame(fix_time=c(87,85.6),a_speed=c(0.609,0.472),b_speed=c(1.715,1.685),separation=c(242.7,243.4),                     s_separation=c(657.5,657.9),rate_terms=c(95,95),                     load_time=c(18.1,18.2),highpressure_time=c(82,60),mpa=c(78,28.8)) predict(m,newdata=new.data,type='class') ##의사결정나무 연습문제 traintxt=read.csv('occupancy_train.csv') library(glmnet) xmat=as.matrix(traintxt[2:6]) #종속변수 제외하고는 매트릭스 형태로 바꿔줌/ 데이터프레임[]는 열을 의미 yvec=traintxt$Occupanc fit.lasso=glmnet(x=xmat,y=yvec,alpha=1,nlambda = 100) #알파 : 비용(1이 제일 큼) fit.lasso.cv=cv.glmnet(x=xmat,y=yvec,nfolds=10,alpha=1,lambda = fit.lasso$lambda) #nfold수행횟수 plot(fit.lasso.cv) fit.lasso.param=fit.lasso.cv$lambda.min #람다 가장 작은 값 fit.lasso.param=fit.lasso.cv$lambda.1se #처음 커지는 람다 값 fit.lasso.tune=glmnet(x=xmat,y=yvec,alpha = 1,lambda=fit.lasso.param) coef(fit.lasso.tune) #계수 #분류나무 t_index=sample(1:nrow(traintxt),size=nrow(traintxt)*0.7) train=traintxt[t_index,] test=traintxt[-t_index,] m=tree(factor(Occupancy)~Temperature+Light+CO2,data=train) plot(m) text(m) #예측값과 정확도 yhat_test=predict(m,test,type='class') table=table(real=test$Occupancy,predict=yhat_test);table (table[1,1]+table[2,2])/sum(table) #ROC AUC library(Epi) ROC(test=yhat_test,stat=test$Occupancy,plot='ROC',AUC=T,main='tree') ##KNN연습문제 occall=read.csv('occupancy_all.csv',header=T) occtrain=read.csv('occupancy_train.csv',header=T) occtest=read.csv('occupancy_test.csv',header=T) t_index=sample(1:nrow(occall),size=nrow(occall*0.7) train=occtrain[t_index,] test=occall[-t_index,] head(train) head(test) xmat.train=as.matrix(occtrain[c(2,4,5)]) occ.train=occtrain$Occupancy xmat.test=as.matrix(occtest[c(2,4,5)]) library(class) head(xmat.train) head(xmat.test) yhat_test=knn(xmat.train,xmat.test,as.factor(occ.train),k=3) table=table(real=occtest$Occupancy,predict=yhat_test);table (table[1,1]+table[2,2])/sum(table) #k값 library(e1071) tune.out=tune.knn(x=xmat.train,y=as.factor(occ.train),k=1:10);tune.out plot(tune.out) #ROC ROC(test=yhat_test,stat=occtest$Occupancy,plot='ROC',AUC=T,main='tree')