DATA MINING
Desktop Survival Guide
by Graham Williams

## Colourful Correlations

You could write your own path.colors as below and obtain a more colourful correlation plot. The colours are quite garish but it gives an idea of what is possible--The reds and purples give a good indication of high correlation (negative and positive), while the blues and greens identify less correlation.

[width=]rplot-corr-wine

 ```# Suggested by Duncan Murdoch path.colors <- function(n, path=c('cyan', 'white', 'magenta'), interp=c('rgb','hsv')) { interp <- match.arg(interp) path <- col2rgb(path) nin <- ncol(path) if (interp == 'hsv') { path <- rgb2hsv(path) # Modify the interpolation so that the circular nature of hue for (i in 2:nin) path[1,i] <- path[1,i] + round(path[1,i-1]-path[1,i]) result <- apply(path, 1, function(x) approx(seq(0, 1, len=nin), x, seq(0, 1, len=n))\$y) return(hsv(result[,1] %% 1, result[,2], result[,3])) } else { result <- apply(path, 1, function(x) approx(seq(0, 1, len=nin), x, seq(0, 1, len=n))\$y) return(rgb(result[,1]/255, result[,2]/255, result[,3]/255)) } } pdf('graphics/rplot-corr-wine.pdf') library(ellipse) load('wine.Rdata') corr.wine <- cor(wine) ord <- order(corr.wine[1,]) xc <- corr.wine[ord, ord] plotcorr(xc, col=path.colors(11, c("red","green", "blue","red"), interp="hsv")[5*xc + 6]) dev.off() ```

http://rattle.togaware.com/code/rplot-corr-wine.R