Send output to:
Browser Blue - Charts White
Browser Black/White
CSV
Data X:
79 30 94 112285 146283 58 28 103 84786 98364 60 38 93 83123 86146 108 30 103 101193 96933 49 22 51 38361 79234 0 26 70 68504 42551 121 25 91 119182 195663 1 18 22 22807 6853 43 26 93 116174 95757 69 25 60 57635 85584 78 38 123 66198 143983 86 44 148 71701 75851 44 30 90 57793 59238 104 40 124 80444 93163 63 34 70 53855 96037 158 47 168 97668 151511 102 30 115 133824 136368 77 31 71 101481 112642 82 23 66 99645 94728 115 36 134 114789 105499 101 36 117 99052 121527 80 30 108 67654 127766 50 25 84 65553 98958 83 39 156 97500 77900 123 34 120 69112 85646 73 31 114 82753 98579 81 31 94 85323 130767 105 33 120 72654 131741 47 25 81 30727 53907 105 33 110 77873 178812 94 35 133 117478 146761 44 42 122 74007 82036 114 43 158 90183 163253 38 30 109 61542 27032 107 33 124 101494 171975 71 32 92 55813 86572 84 36 126 79215 159676 59 28 70 55461 85371 33 14 37 31081 58391 96 32 120 83122 136815 106 30 93 70106 120642 56 35 95 60578 69107 59 28 90 79892 108016 39 28 80 49810 46341 34 39 31 71570 78348 76 34 110 100708 79336 20 26 66 33032 56968 91 39 138 82875 93176 115 39 133 139077 161632 85 33 113 71595 87850 76 28 100 72260 127969 8 4 7 5950 15049 79 39 140 115762 155135 21 18 61 32551 25109 30 14 41 31701 45824 76 29 96 80670 102996 101 44 164 143558 160604 92 28 102 120733 162647 123 35 124 105195 174141 75 28 99 73107 60622 128 38 129 132068 179566 105 23 62 149193 184301 55 36 73 46821 75661 56 32 114 87011 96144 41 29 99 95260 129847 72 25 70 55183 117286 67 27 104 106671 71180 75 36 116 73511 109377 114 28 91 92945 85298 118 23 74 78664 73631 77 40 138 70054 86767 22 23 67 22618 23824 66 40 151 74011 93487 69 28 72 83737 82981 105 34 120 69094 73815 116 33 115 93133 94552 88 28 105 95536 132190 73 34 104 225920 128754 99 30 108 62133 66363 62 33 98 61370 67808 53 22 69 43836 61724 118 38 111 106117 131722 30 26 99 38692 68580 100 35 71 84651 106175 49 8 27 56622 55792 24 24 69 15986 25157 67 29 107 95364 76669 57 29 107 89691 105805 75 45 93 67267 129484 135 37 129 126846 72413 68 33 69 41140 87831 124 33 118 102860 96971 33 25 73 51715 71299 98 32 119 55801 77494 58 29 104 111813 120336 68 28 107 120293 93913 81 28 99 138599 136048 131 31 90 161647 181248 110 52 197 115929 146123 130 24 85 162901 186646 93 41 139 109825 102255 118 33 106 129838 168237 39 32 50 37510 64219 13 19 64 43750 19630 74 20 31 40652 76825 81 31 63 87771 115338 109 31 92 85872 109427 151 32 106 89275 118168 28 23 69 192565 153197 83 30 93 140867 68370 54 31 114 120662 146304 133 42 110 101338 103950 12 1 0 1168 5841 106 32 83 65567 84396 23 11 30 25162 24610 71 36 98 40735 55515 116 31 82 91413 209056 4 0 0 855 6622 62 24 60 97068 115814 18 8 9 14116 13155 98 33 115 76643 142775 64 40 140 110681 68847 32 38 120 92696 20112 25 24 66 94785 61023 16 8 21 8773 13983 48 35 124 83209 65176 100 43 152 93815 132432 46 43 139 86687 112494 129 41 144 105547 170875 130 38 120 103487 180759 136 45 160 213688 214921 59 31 114 71220 100226 32 28 78 56926 54454 63 31 119 91721 78876 95 40 141 115168 170745 14 30 101 111194 6940 113 37 133 135777 122037 47 30 83 51513 53782 92 35 116 74163 127748 70 32 90 51633 86839 19 27 36 75345 44830 91 31 97 98952 103300 111 31 98 102372 112283 41 21 78 37238 10901 120 39 117 103772 120691 135 41 148 123969 58106 87 32 105 135400 122422 131 39 132 130115 139296 4 0 0 6023 7953 47 30 73 64466 89455 109 37 86 54990 147866 7 0 0 1644 4245 12 5 13 6179 21509 0 1 4 3926 7670 37 32 48 34777 14336 46 24 46 73224 53608
Names of X columns:
blogged_computations compendiums_reviewed feedback_messages_p120 totsize totseconds
Endogenous Variable (Column Number)
Categorization
none
quantiles
hclust
equal
Number of categories (only if categorization<>none)
Cross-Validation? (only if categorization<>none)
no
yes
Chart options
R Code
library(party) library(Hmisc) par1 <- as.numeric(par1) par3 <- as.numeric(par3) x <- data.frame(t(y)) is.data.frame(x) x <- x[!is.na(x[,par1]),] k <- length(x[1,]) n <- length(x[,1]) colnames(x)[par1] x[,par1] if (par2 == 'kmeans') { cl <- kmeans(x[,par1], par3) print(cl) clm <- matrix(cbind(cl$centers,1:par3),ncol=2) clm <- clm[sort.list(clm[,1]),] for (i in 1:par3) { cl$cluster[cl$cluster==clm[i,2]] <- paste('C',i,sep='') } cl$cluster <- as.factor(cl$cluster) print(cl$cluster) x[,par1] <- cl$cluster } if (par2 == 'quantiles') { x[,par1] <- cut2(x[,par1],g=par3) } if (par2 == 'hclust') { hc <- hclust(dist(x[,par1])^2, 'cen') print(hc) memb <- cutree(hc, k = par3) dum <- c(mean(x[memb==1,par1])) for (i in 2:par3) { dum <- c(dum, mean(x[memb==i,par1])) } hcm <- matrix(cbind(dum,1:par3),ncol=2) hcm <- hcm[sort.list(hcm[,1]),] for (i in 1:par3) { memb[memb==hcm[i,2]] <- paste('C',i,sep='') } memb <- as.factor(memb) print(memb) x[,par1] <- memb } if (par2=='equal') { ed <- cut(as.numeric(x[,par1]),par3,labels=paste('C',1:par3,sep='')) x[,par1] <- as.factor(ed) } table(x[,par1]) colnames(x) colnames(x)[par1] x[,par1] if (par2 == 'none') { m <- ctree(as.formula(paste(colnames(x)[par1],' ~ .',sep='')),data = x) } load(file='createtable') if (par2 != 'none') { m <- ctree(as.formula(paste('as.factor(',colnames(x)[par1],') ~ .',sep='')),data = x) if (par4=='yes') { a<-table.start() a<-table.row.start(a) a<-table.element(a,'10-Fold Cross Validation',3+2*par3,TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'',1,TRUE) a<-table.element(a,'Prediction (training)',par3+1,TRUE) a<-table.element(a,'Prediction (testing)',par3+1,TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'Actual',1,TRUE) for (jjj in 1:par3) a<-table.element(a,paste('C',jjj,sep=''),1,TRUE) a<-table.element(a,'CV',1,TRUE) for (jjj in 1:par3) a<-table.element(a,paste('C',jjj,sep=''),1,TRUE) a<-table.element(a,'CV',1,TRUE) a<-table.row.end(a) for (i in 1:10) { ind <- sample(2, nrow(x), replace=T, prob=c(0.9,0.1)) m.ct <- ctree(as.formula(paste('as.factor(',colnames(x)[par1],') ~ .',sep='')),data =x[ind==1,]) if (i==1) { m.ct.i.pred <- predict(m.ct, newdata=x[ind==1,]) m.ct.i.actu <- x[ind==1,par1] m.ct.x.pred <- predict(m.ct, newdata=x[ind==2,]) m.ct.x.actu <- x[ind==2,par1] } else { m.ct.i.pred <- c(m.ct.i.pred,predict(m.ct, newdata=x[ind==1,])) m.ct.i.actu <- c(m.ct.i.actu,x[ind==1,par1]) m.ct.x.pred <- c(m.ct.x.pred,predict(m.ct, newdata=x[ind==2,])) m.ct.x.actu <- c(m.ct.x.actu,x[ind==2,par1]) } } print(m.ct.i.tab <- table(m.ct.i.actu,m.ct.i.pred)) numer <- 0 for (i in 1:par3) { print(m.ct.i.tab[i,i] / sum(m.ct.i.tab[i,])) numer <- numer + m.ct.i.tab[i,i] } print(m.ct.i.cp <- numer / sum(m.ct.i.tab)) print(m.ct.x.tab <- table(m.ct.x.actu,m.ct.x.pred)) numer <- 0 for (i in 1:par3) { print(m.ct.x.tab[i,i] / sum(m.ct.x.tab[i,])) numer <- numer + m.ct.x.tab[i,i] } print(m.ct.x.cp <- numer / sum(m.ct.x.tab)) for (i in 1:par3) { a<-table.row.start(a) a<-table.element(a,paste('C',i,sep=''),1,TRUE) for (jjj in 1:par3) a<-table.element(a,m.ct.i.tab[i,jjj]) a<-table.element(a,round(m.ct.i.tab[i,i]/sum(m.ct.i.tab[i,]),4)) for (jjj in 1:par3) a<-table.element(a,m.ct.x.tab[i,jjj]) a<-table.element(a,round(m.ct.x.tab[i,i]/sum(m.ct.x.tab[i,]),4)) a<-table.row.end(a) } a<-table.row.start(a) a<-table.element(a,'Overall',1,TRUE) for (jjj in 1:par3) a<-table.element(a,'-') a<-table.element(a,round(m.ct.i.cp,4)) for (jjj in 1:par3) a<-table.element(a,'-') a<-table.element(a,round(m.ct.x.cp,4)) a<-table.row.end(a) a<-table.end(a) table.save(a,file='mytable3.tab') } } m bitmap(file='test1.png') plot(m) dev.off() bitmap(file='test1a.png') plot(x[,par1] ~ as.factor(where(m)),main='Response by Terminal Node',xlab='Terminal Node',ylab='Response') dev.off() if (par2 == 'none') { forec <- predict(m) result <- as.data.frame(cbind(x[,par1],forec,x[,par1]-forec)) colnames(result) <- c('Actuals','Forecasts','Residuals') print(result) } if (par2 != 'none') { print(cbind(as.factor(x[,par1]),predict(m))) myt <- table(as.factor(x[,par1]),predict(m)) print(myt) } bitmap(file='test2.png') if(par2=='none') { op <- par(mfrow=c(2,2)) plot(density(result$Actuals),main='Kernel Density Plot of Actuals') plot(density(result$Residuals),main='Kernel Density Plot of Residuals') plot(result$Forecasts,result$Actuals,main='Actuals versus Predictions',xlab='Predictions',ylab='Actuals') plot(density(result$Forecasts),main='Kernel Density Plot of Predictions') par(op) } if(par2!='none') { plot(myt,main='Confusion Matrix',xlab='Actual',ylab='Predicted') } dev.off() if (par2 == 'none') { detcoef <- cor(result$Forecasts,result$Actuals) a<-table.start() a<-table.row.start(a) a<-table.element(a,'Goodness of Fit',2,TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'Correlation',1,TRUE) a<-table.element(a,round(detcoef,4)) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'R-squared',1,TRUE) a<-table.element(a,round(detcoef*detcoef,4)) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'RMSE',1,TRUE) a<-table.element(a,round(sqrt(mean((result$Residuals)^2)),4)) a<-table.row.end(a) a<-table.end(a) table.save(a,file='mytable1.tab') a<-table.start() a<-table.row.start(a) a<-table.element(a,'Actuals, Predictions, and Residuals',4,TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'#',header=TRUE) a<-table.element(a,'Actuals',header=TRUE) a<-table.element(a,'Forecasts',header=TRUE) a<-table.element(a,'Residuals',header=TRUE) a<-table.row.end(a) for (i in 1:length(result$Actuals)) { a<-table.row.start(a) a<-table.element(a,i,header=TRUE) a<-table.element(a,result$Actuals[i]) a<-table.element(a,result$Forecasts[i]) a<-table.element(a,result$Residuals[i]) a<-table.row.end(a) } a<-table.end(a) table.save(a,file='mytable.tab') } if (par2 != 'none') { a<-table.start() a<-table.row.start(a) a<-table.element(a,'Confusion Matrix (predicted in columns / actuals in rows)',par3+1,TRUE) a<-table.row.end(a) a<-table.row.start(a) a<-table.element(a,'',1,TRUE) for (i in 1:par3) { a<-table.element(a,paste('C',i,sep=''),1,TRUE) } a<-table.row.end(a) for (i in 1:par3) { a<-table.row.start(a) a<-table.element(a,paste('C',i,sep=''),1,TRUE) for (j in 1:par3) { a<-table.element(a,myt[i,j]) } a<-table.row.end(a) } a<-table.end(a) table.save(a,file='mytable2.tab') }
Compute
Summary of computational transaction
Raw Input
view raw input (R code)
Raw Output
view raw output of R engine
Computing time
2 seconds
R Server
Big Analytics Cloud Computing Center
Click here to blog (archive) this computation