R

• RStudio
• Emacs speaks statistics
• RGedit

• R Manuals
• Bent Jorgensen's exhaustive Cheat Sheet
• My Notes
• R Style Guide
• F-Race

How to find out about stuff in R?

> ?function > help.search("search string") > RSiteSearch("search string") > names(object) or attributes(object) > str(object) > kmeans function (x, centers, iter.max = 10, nstart = 1, algorithm = c("Hartigan-Wong", "Lloyd", "Forgy", "MacQueen")) { ...

http://rseek.org/

The bash script to parse the data

#!/bin/bash for a in *.tgz ; do tar xzvf $a; done for a in ??????; do for r in $a/res/*; do echo $a `basename $r` `cat $r`; done; done > results.txt

A script preparing the data for the analysis:

## we load the data and give names to fields D <- read.table("results2.txt",fill=TRUE) colnames(D)<- c("alg","instance","col","time") ## we handle some errors in the data D$instance<-gsub("queens([0-9]+).*","queen\\1_\\1.txt",D$instance) D$instance<-factor(D$instance) levels(D$instance) str(D) D$instance<-gsub("queen([0-9]+)\\.txt","queen\\1_\\1.txt",D$instance) D$instance<-factor(D$instance) levels(D$instance) str(D) D$size<-(as.numeric(gsub("queen([0-9]+)_.*","\\1",D$instance)))^2 ## we check everybody has the same number of runs table(D$alg,D$instance) ## we found some instances that were not solved by everybody, hence we remove them D1<-D[!D$instance %in% c("queen1_1.txt","queen2_2.txt","queen3_3.txt"),] ## algorithms as well as instances must be treated as factors D1$alg<-factor(D1$alg) D1$instance<-factor(D1$instance) ## we add a column with the size of the instance ## we store the data in perfect shape for future analysis DATA<-D1 save(DATA,file="results.rda")

The analysis of results:

## load the data load("results.rda") ## inspect them str(DATA) ## also visually bwplot(reorder(alg, col, median)~col,data=DATA) bwplot(reorder(alg, col, median)~col|instance,data=DATA) ## Let's make the ranks of the colors T1 <- split(DATA["col"], DATA["instance"]) T2 <- lapply(T1, rank, na.last = "keep") T3 <- unsplit(T2, DATA["instance"]) DATA$rank <- T3 ## we plot the ranks for an aggregate analysis ## reoder sort the factor algorithm by median values bwplot(reorder(alg, rank, median) ~ rank, data = DATA) ## let's analyse the time ## this plot is not very helpful xyplot(col~time|instance,groups=alg,data=DATA) ## another visualization dotplot(instance~time | alg,data=DATA) ## inspect the growth xyplot(time~size | alg,data=DATA,type=c("p", "a"),scales=list(x=list(log=FALSE),y=list(log=FALSE,relation="same"))) xyplot(time~size | alg,data=DATA,type=c("p", "a"),scales=list(x=list(log=TRUE),y=list(log=TRUE,relation="same"))) ## we cannot aggregate time over instances due to differences in scales ## let's then prepare the ranks also for the time T1 <- split(DATA["time"], DATA["instance"]) T2 <- lapply(T1, rank, na.last = "keep") T3 <- unsplit(T2, DATA["instance"]) DATA$rank.time <- T3 ## xyplot(rank.time ~ rank, data = DATA,groups = alg) ## we prepare the analysis on median values only: a<-aggregate(DATA$rank,list(DATA$alg),median) b<-aggregate(DATA$rank.time,list(DATA$alg),median) c<-merge(a,b,by="Group.1") colnames(c)<-c("alg","rank","rank.time") xyplot(rank.time ~ rank, data = c, groups = alg, ##scales=list(y=list(log=TRUE,relation="free")), panel = function(x, y, subscripts, groups) ltext(x = x, y = y, labels = groups[subscripts], cex=1, fontfamily = "HersheySans"))