Send output to:
Browser Blue - Charts White
Browser Black/White
CSV
Data:
280190 280408 276836 275216 274352 271311 289802 290726 292300 278506 269826 265861 269034 264176 255198 253353 246057 235372 258556 260993 254663 250643 243422 247105 248541 245039 237080 237085 225554 226839 247934 248333 246969 245098 246263 255765 264319 268347 273046 273963 267430 271993 292710 295881 293299 288576 286445 297584 300431 298522 292213 285383 277537 277891 302686 300653 296369 287224 279998 283495 285775 282329 277799 271980 266730 262433 285378 286692 282917 277686 274371 277466 290604 290770 283654 278601 274405 272817 294292 300562 298982 296917 295008 297295
Sample Range:
(leave blank to include all observations)
From:
To:
# blockwidth
Chart options
R Code
par1 <- as.numeric(par1) (n <- length(x)) (np <- floor(n / par1)) arr <- array(NA,dim=c(par1,np+1)) darr <- array(NA,dim=c(par1,np+1)) ari <- array(0,dim=par1) dx <- diff(x) j <- 0 for (i in 1:n) { j = j + 1 ari[j] = ari[j] + 1 arr[j,ari[j]] <- x[i] darr[j,ari[j]] <- dx[i] if (j == par1) j = 0 } ari arr darr arr.mean <- array(NA,dim=par1) arr.median <- array(NA,dim=par1) arr.midrange <- array(NA,dim=par1) for (j in 1:par1) { arr.mean[j] <- mean(arr[j,],na.rm=TRUE) arr.median[j] <- median(arr[j,],na.rm=TRUE) arr.midrange[j] <- (quantile(arr[j,],0.75,na.rm=TRUE) + quantile(arr[j,],0.25,na.rm=TRUE)) / 2 } overall.mean <- mean(x) overall.median <- median(x) overall.midrange <- (quantile(x,0.75) + quantile(x,0.25)) / 2 bitmap(file='plot1.png') plot(arr.mean,type='b',ylab='mean',main='Mean Plot',xlab='Periodic Index') mtext(paste('#blocks = ',np)) abline(overall.mean,0) dev.off() bitmap(file='plot2.png') plot(arr.median,type='b',ylab='median',main='Median Plot',xlab='Periodic Index') mtext(paste('#blocks = ',np)) abline(overall.median,0) dev.off() bitmap(file='plot3.png') plot(arr.midrange,type='b',ylab='midrange',main='Midrange Plot',xlab='Periodic Index') mtext(paste('#blocks = ',np)) abline(overall.midrange,0) dev.off() bitmap(file='plot4.png') z <- data.frame(t(arr)) names(z) <- c(1:par1) (boxplot(z,notch=TRUE,col='grey',xlab='Periodic Index',ylab='Value',main='Notched Box Plots - Periodic Subseries')) dev.off() bitmap(file='plot4b.png') z <- data.frame(t(darr)) names(z) <- c(1:par1) (boxplot(z,notch=TRUE,col='grey',xlab='Periodic Index',ylab='Value',main='Notched Box Plots - Differenced Periodic Subseries')) dev.off() bitmap(file='plot5.png') z <- data.frame(arr) names(z) <- c(1:np) (boxplot(z,notch=TRUE,col='grey',xlab='Block Index',ylab='Value',main='Notched Box Plots - Sequential Blocks')) dev.off() bitmap(file='plot6.png') z <- data.frame(cbind(arr.mean,arr.median,arr.midrange)) names(z) <- list('mean','median','midrange') (boxplot(z,notch=TRUE,col='grey',ylab='Overall Central Tendency',main='Notched Box Plots')) dev.off()
Compute
Summary of computational transaction
Raw Input
view raw input (R code)
Raw Output
view raw output of R engine
Computing time
0 seconds
R Server
Big Analytics Cloud Computing Center
Click here to blog (archive) this computation