panel.first in formula plot revisited

Today I scouted an issue, which I already have solved at least twice formerly - and couldn't remember at all. Getting older and older... 　

d.my <- data.frame(cal_d=as.Date(c("2011-12-30","2012-01-03","2012-04-07","2012-07-08")), y=1:4) 　

The following code would fail to produce a grid():
plot(y ~ cal_d, d.my, panel.first=grid())

This will paint the grid, but choose the wrong ticks:
plot(y ~ cal_d, d.my, panel.first=quote(grid()))

This will paint the grid correctly:
plot(y ~ cal_d, d.my, panel.first=quote({
    grid(nx=NA, ny=NULL)
    abline(v=axis.Date(1, x=d.my$cal_d, labels=NA, col=NA), col="grey", lty="dotted")
})) 　

Reference found: https://stat.ethz.ch/pipermail/r-help/2011-May/279169.html

Methods and Classes

How can we find the defined functions for a special class?

# for which classes does my function provide an own interface?
methods("summary")

# which functions do support my class?
methods(class="matrix")

# get the source code, if not readily available
getAnywhere("Desc.character")

DescTools approaching v. 1.0...

A next version of DescTools 0.99.16 has been submitted to CRAN today.
It includes a handful new functions and several performance enhancements.
See: http://cran.r-project.org/web/packages/DescTools/NEWS

p-values uniformly distributed

Are p-values really uniformly distributed? They are:


library(DescTools)
gg <- rnorm(10000)
n <- 100
m <- sapply(1:n, function(x) sample(1:10000, size = n))
m[] <- gg[m]
d.frm <- data.frame(x=as.vector(m), grp=factor(rep(1:n, each=n)))
p <- na.omit(as.vector(pairwise.t.test(x = d.frm$x, g = d.frm$grp,
                                       p.adjust.method = "none", pool.sd = FALSE)$p.value))
Desc(p)


See: https://www.youtube.com/watch?v=5OL1RqHrZQ8

Unique List, but with order

How to create a list of elements in order:

(x <- structure(c(3L, 3L, 3L, 3L, 1L,  2L, 2L,
            2L, 2L, 3L, 2L, 1L, 1L, 2L, 2L,
            2L, 2L, 2L, 2L, 2L, 2L, 2L),
            .Label = c("B", "C", "D"), class = "factor"))
[1] D D D D B C C C C D C B B C C C C C C C C C
Levels: B C D

> x[c(TRUE, diff(as.numeric(x))!=0)]
[1] D B C D C B C
Levels: B C D

 

Elegant Summary

A really elegant solution for combining different summary results:

library(plyr)
mtcars$mpg_g <- cut(mtcars$mpg, breaks=seq(from=10, to=35, by=5))
d.res <- ddply(mtcars, c("cyl", "am"), summarize,
           anz = length(cyl),
           min.hp=min(hp), med.hp=median(hp), mean.hp=round(mean(hp),2), max.hp=max(hp),
           mpg_g=matrix(table(mpg_g), nrow=1))
d.res

Credits to Markus Naepflin (2014) for that.

DescTools 0.99.6 released

The first Version of DescTools can be downloaded from CRAN. So it is available as well in Mac-Version:
http://cran.r-project.org/web/packages/DescTools/index.html

DescTools contains a bunch of basic statistic functions and convenience wrappers for efficiently describing data, creating specific plots, doing reports using MS Word, Excel or PowerPoint. The package's intention is to offer a toolbox, which facilitates the (notoriously time consuming) first descriptive tasks in data analysis, consisting of calculating descriptive statistics, drawing graphical summaries and reporting the results. Many of the included functions can be found scattered in other packages and other sources written partly by Titans of R. The reason for collecting them here, was primarily to have them consolidated in ONE instead of dozens of packages (which themselves might depend on other packages which are not needed at all), and to provide a common and consistent interface as far as function and arguments naming, NA handling, recycling rules etc. are concerned. Google style guides were used as naming rules (in absence of convincing alternatives). The 'camel style' was consequently applied to functions borrowed from contributed R packages as well.

Wordcloud

This code yields interesting insights:

library(wordcloud)
txt <- c("hrcds_b","amb_spit_nvl_b","hpp_fg","spez_kons_n","treu_fg",
         "qz_n","notfall_n","stat_pv_bl_b","vp_at_p","hosp_n","weibl_p")
n <- c(0.247,0.167,0.084,0.04,
       -0.009,-0.012,-0.062,-0.14,-0.143,-0.229,-0.363)
txt <- Sort(data.frame(word=txt, freq=abs(n*1000),
                       col=PalHelsana()[c("rot", "hellblau")][(n>0)*1+1],
                       stringsAsFactors=FALSE),2, decr=TRUE)
wordcloud(txt$word, txt$freq, colors=txt$col, ordered.colors=TRUE)

How to operate in parallel

Doing parallel calculations will spare you much time... There are three points deserving our attention. %dopar% does parallel calculation, the combining function after having the work done can be defined, either c or rbind or whatever and - last but not least - the packages used within the loop must be defined in the specific parameter.

library(doParallel)
cl <- makeCluster(3)   # the number of cores to be used
registerDoParallel(cl)
getDoParWorkers()      # are they ready?

# remind defining packages if they're used within the loop
res <- foreach(i=1:5, .combine=c, .packages="DescTools") %dopar% {
  Primes(i)
} 
res
stopCluster(cl)       # release your slaves again


Boot already has native support for parallel working:

library(boot)
slopeFun <- function(df, i) {
  #df must be a data frame. 
  #i is the vector of row indices that boot will pass 
  xResamp <- df[i, ]
  slope <- lm(hp ~ cyl + disp, data=xResamp)$coef[2]
}

ptime <- system.time({
  b <- boot(mtcars, slopeFun, R=50000, ncpus=6, parallel="snow")
})[3]
ptime

Tables in R

We have quite a few table functions in R, all of them solving a specific problem.

library(DescTools)
# Let's define a 3-dim table
# source: http://www.math.wpi.edu/saspdf/stat/chap28.pdf, page 1248
school <- as.table(array(c(35,29, 14,27, 32,10, 53,23), dim=c(2,2,2),
                   dimnames=list(enrollment=c("yes","no"),
                                 internship=c("yes","no"), gender=c("boys","girls"))))
school

# the percentages
prop.table(school)

# printed as flat table
ftable(school, row.vars=c(3,2))

# aggregate and layout
xtabs(Freq ~ enrollment + gender, data=school)

# and the inverse Operation
tab <- xtabs(Freq ~ enrollment + gender, data=school)
d.frm <- as.data.frame.table(tab)

# factor form with frequencies
data.frame(school)

# get the raw data back in a data.frame
head(Untable(school))

# describe with bells and whistles
Desc(xtabs(Freq ~ internship + enrollment, data=school) )
Assocs(xtabs(Freq ~ internship + enrollment, data=school) )

# how to make a data.frame out of a table
tab <- table(d.pizza$driver, d.pizza$area)
d.frm <- data.frame(unclass(tab))

# or even simpler with base function:
d.frm <- as.data.frame.matrix(tab)

# combine several arrays:
abs <- apply(Titanic, 1:length(dim(Titanic)), formatC, width=3)
rel <- apply(prop.table(Titanic), 1:length(dim(Titanic)), FormatFix, after=3)
ftable(abind(abs, rel))

Plot intersecting areas

Ever had to plot intersecting areas?
From now on I would use rgeos to do similar things. The following does the trick:

library(rgeos)
library(DescTools)

Canvas(, xpd=TRUE)
p1 <- as(DrawCircle(0.5)[[1]], "gpc.poly")
p2 <- as(DrawCircle(-0.5,0.5)[[1]], "gpc.poly")
p3 <- as(DrawCircle(0,-0.5)[[1]], "gpc.poly")
plot(append.poly(append.poly(p1, p2), p3), axes=FALSE, frame.plot=FALSE, xlab="", ylab="")
plot(intersect(p1, p2), poly.args = list(col = "red"), add = TRUE)
plot(intersect(intersect(p1, p2), p3), poly.args = list(col = "blue"), add = TRUE)

Second axis

Just a short snippet for a plot with 2 axes for demonstrating the technique.


b <- barplot(1:10, ylim=c(0,20))

par(new=TRUE)
plot(x=b, y=rep(60, 10), xlim=par("usr")[1:2], xaxs="i", yaxs="i"
     , frame.plot=FALSE, axes=FALSE, xlab="", ylab="", type="b", pch=15
     , ylim=c(0,80))

axis(side=4)


And the result:

Barplot in Tibco Style

Tibco has a nice software which produces good looking plots:

tibco.colors <- apply( mcol <- matrix(c(
  0,91,0, 0,157,69, 253,1,97, 60,120,177,
  156,205,36, 244,198,7, 254,130,1,
  96,138,138, 178,113,60
), ncol=3, byrow=TRUE), 1, function(x) rgb(x[1], x[2], x[3], max=255))

x <- c(12,8,6,4,3,1,1)

par(mgp=c(2.2,0.7,0) )     # decrease distance label to axis

b <- barplot( x, col=rev(tibco.colors), main="Tibco style", border=NA, cex.axis=0.7, las=1, ylim=c(0,13), yaxt="n")

text( b, x+0.6, label=x, cex=0.7)

axis(side=1, at=b, labels=c("EX","VIT","EC","CE","LA","CO","PA"), cex.axis=0.7, las=2, col="grey", las=1, tck=0, xaxs="i")

abline(h=0, col="grey")    # just to ensure the axis going through (0,0)

# find the centers of the bars and the gaps
run.mean <- filter( b, filter=c(0.5,0.5))[-length(b)]
gapx <- c(run.mean[1]-diff(b)[1], run.mean, run.mean+diff(b) )
rug(gapx, -0.025, col="grey") 

axis(side=2, at=0:12, labels=0:12, cex.axis=0.7, las=2, col="grey", las=1, tck=-0.025)

legend(x="topright", legend=c("Marking","","CONMEDS","DEMOGRAPHICS","ECG","EXPOSURE","LAB_RESULT")
  , fill=c(tibco.colors[1],NA,tibco.colors[2:10]), cex=0.7, box.col=NA, border=c("black",NA,rep("black",10)))

dots arguments

How can we parse the dots-arguments in a function?

dots <- function(...){
# was cex in the dots-args? parse dots.arguments
cex <- unlist(match.call(expand.dots=FALSE)$...["cex"])
if(is.null(cex)) print("No cex supplied")
else print(gettextf("cex was %s", cex))
}

dots (cex=0.9, another=4)
dots (another=5)

Get all loaded data.frames

Which data.frames have we loaded?
ls()[ lapply( lapply(ls(), function(x) gettextf("class(%s)", x)), function(x) eval(parse(text=x))) == "data.frame" ]

All available functions in a package:
ls.str(pos = 1, mode = "function")

All installed packages:
search()
(.packages())
(installed.packages())

All available function on search() path:
strsplit( lsf.str(pos = length(search())), " :")

Conditional Mosaics: Cotabplot

Fine implementation, thanks to Achim Zeileis.

library(vcd)
data("Punishment", package = "vcd")
pun <- xtabs(Freq ~ memory + attitude + age + education, data = Punishment)
cotabplot(~ memory + attitude | age + education, data = pun, panel = cotab_coindep,
  n = 5000, type = "mosaic", test = "maxchisq", interpolate = 1:2 )

Conversions

How can we split a matrix columnwise to a list?
m <- matrix(rnorm(10 * 2), ncol = 2)

The first solution uses a double conversion via a data.frame:
as.list(as.data.frame(m))

The second uses split:
split(x=t(m), f=1:ncol(m))

The third solution works with lapply:
lapply( seq_len(ncol(m)), function(k) m[,k])

Flexible arguments for R functions

How can we remain flexible concerning arguments of an R function used within a user function. Say we want a histogram to be included in our plot and be able to define all of it's arguments.
I found a good solution in some R-Core-code (don't remember exactly where...).
It goes like that:

myFunction <- function( x, main = "", args.hist = NULL, ... ) {

  # define default arguments
  args.hist1 <- list(x = x, xlab = "", ylab = "", freq = FALSE,
      xaxt = "n", xlim = NULL, ylim = NULL, main = main, las = 1,
      col =  "white", cex.axis = 1.2)
  # override default arguments with user defined ones
  if (!is.null(args.hist)) {
    args.hist1[names(args.hist)] <- args.hist
  }
  # call function by means of do.call with the list of arguments
  # we can even filter the arguments by name here...
  res <- do.call("hist", c(args.hist1[names(args.hist1) %in%
      c("x", "breaks", "include.lowest", "right", "nclass")],
      plot = FALSE))
}

The function call would then be:
  myFunction( x=1:10, args.hist = list(right = TRUE) )

So we have a fine control over which arguments we want to be passed (include.lowest, right) and which ones not (ex. col, cex.axis).

Find all subsets

How can we get all subsets of a vector's elements?

all.subsets <- function(x, min.n=1, max.n=length(x)){
  # return a list with all subsets of x
  # Caution: This can be heavy for moderate lengths of x11
  lst <- lapply( min.n:max.n, function(i) {
    m <- combn(x,i) # this is a matrix, split into it's columns
    lapply( seq_len(ncol(m)), function(k) m[,k])
  } )
  # Alternative:
  # lst <- lapply(min.n:max.n, function(i) lapply(apply(combn(x,i),2,list),unlist))

  # and now flatten the list of lists into one list
  lst <- split(unlist(lst), rep(1:length(idx <- rapply(lst, length)), idx))
  return(lst)
}

# example:
y <- letters[1:5]
all.subsets(y)

Run all examples of a package

Get all the base and recommended packages:
installed.packages(priority = c("base","recommended"))

A list of the functions in a package can easily be created by:
ls(pos = "package:MASS")

Run the examples for the first 10 functions of the package MASS:
for( x in paste("example(`", ls(pos="package:MASS")[1:10],"`)", sep="")){
  eval(parse(text=x))
}

Running mean, median and others

The running mean (moving average) can be calculated in R by means of the function
x <- c(1,2,4,2,3,4,2,3)
filter(x, rep(1/3,3) )

and the running median by
runmed(x,3)

sequential differences by
diff(x)

for other statistics use the library zoo (and there maybe rollapply).
The time series functions can also be useful for similar problems (e.g. deltat, cycle).

package {caTools} has implemented some fast algorithms for similar purposes:
runmean(x), rumin(x), runmax(x), runquantile(x), runmad(x), runsd(x)

regexpr examples (just to remember)

Find names ending in "_id" or in "_c" (this would be the same as x like any ('%_id', '%_c') in SQL flavour):

x <- c("vp_id","man_id","min_A_d","min_B_d","count_n",
       "type_c","birth_d","gender_c","age_n","hist_y")

x[grep("_(id)|c$", x)]
[1] "vp_id" "man_id" "type_c" "gender_c"


Get rid of everything but the digits:
x <- c("485.2.362.q", "222-445", "889 99 8")
gsub(pattern="[[:punct:]]|[[:alpha:]]|[[:blank:]]", replacement="", x)

gsub(pattern="[^0-9]", x=x, replacement="")

gsub(pattern="[^[:digit:]]", x=x, replacement="")


Extract uppercase words from the beginning of a string following the idea "delete everything which is not uppercase words":

x <- c("RONALD AYLMER Fisher", "CHARLES Pearson", "John Tukey")
sapply(x, function(x) StrTrim(sub(
       pattern=sub(pattern="^[A-ZÄÜÖ -]+\\b\\W+\\b", repl="", x=x)
       , repl="", x, fixed=TRUE)))


... and the fine link:
http://www.powerbasic.com/support/help/pbcc/regexpr_statement.htm

Swap variables

... just because of it's elegance:
x <- 2; y <- 5
c(x, y)

y <- (x + y) - (x <- y)
c(x, y)

Recode a factor

Recoding a factor seems not to be a frequent activity of Ripley & Co. as it is astonishingly laboriously (because of the "else" part of it...).
The shortest trick I found so far is:

x <- factor(sample(letters, 20))
y <- x

levels(y) <- list(
  "good"=c("a","b","c"),
  "bad"=c("d","e","f"),
  "ugly"=c("h","m"),
  "else"=setdiff( levels(y), c("a","b","c","d","e","f","h","m") )
)
data.frame(x,y)

Anyone out there a better idea?

Evaluation of text

Just to remember (once for all):
eval( parse( text="a <- c(1,2,3)" ) )

Mengen-Vergleiche

Interessanten Mengenvergleichs-Funktionen:

(A <- c(sort(sample(1:20, 9)),NA))

[1] 1 2 3 12 16 17 18 19 20 NA
(B <- c(sort(sample(3:23, 7)),NA))
[1] 8 10 15 16 18 21 22 NA


# alle Elemente aus A und B (KEINE Duplikate)
union(A, B)
[1] 1 2 3 12 16 17 18 19 20 NA 8 10 15 21 22

# alle Elemente die in A und in B vorkommen (Schnittmenge)
intersect(A, B)
[1] 16 18 NA

# Elemente in A, aber nicht in B vorkommen
setdiff(A, B)
[1] 1 2 3 12 17 19 20

# Elemente in B, aber nicht in A vorkommen
setdiff(B, A)
[1] 8 10 15 21 22


# Enthält A die gleichen Elemente wie B?
setequal(A, B)
[1] FALSE

Function call with a list of arguments

Use do.call to call a function with a previously built list of arguments:

# Define the arguments

args.legend <- list(
  x = "topleft"
  , legend = c(1,2)
  , fill = c("red","blue")
  , xjust = 1, yjust = 1)

plot(1:10, 1:10)
# call function with the arguments' list
do.call("legend", args.legend)

Get information about invisible built-in functions

How can we see the code of invisible functions, i.e. mosaicplot? 

> mosaicplot

yields only:

function (x, ...)
UseMethod("mosaicplot")


but the functions "methods" and "getAnywhere" do the trick:

> methods(mosaicplot)

[1] mosaicplot.default* mosaicplot.formula*

Non-visible functions are asterisked

> getAnywhere("mosaicplot.formula")
A single object matching ‘mosaicplot.formula’ was found
It was found in the following places
registered S3 method for mosaicplot from namespace graphics
namespace:graphics
with value

function (formula, data = NULL, ..., main = deparse(substitute(data)),
subset, na.action = stats::na.omit)
{
main
m <- match.call(expand.dots = FALSE)
.....

Layout-Muster

Split screens:

m <- matrix(0:3, 2, 2)
layout(m, c(1, 3), c(1, 3))
layout.show(3)

m <- matrix(1:4, 2, 2)
layout(m, widths=c(1, 3), heights=c(3, 1))
layout.show(4)

Date formats and functions

Date formats:

Remember the functions:

d <- as.Date(c("1937-01-10","1916-03-02","1913-09-19","1927-12-23","1947-07-28"))

quarters(d)
[1] "Q1" "Q1" "Q3" "Q4" "Q3"


# ... or as function
quarter <- function (x) {
  # Berechnet das Quartal eines Datums
  y <- as.numeric( format( x, "%Y") )
  paste(y, "Q", (as.POSIXlt(x)$mon)%/%3 + 1, sep = "")
}

# ... or alternatively with cut
cut(d, breaks="quarters")
[1] 1937-01-01 1916-01-01 1913-07-01 1927-10-01 1947-07-01
137 Levels: 1913-07-01 1913-10-01 1914-01-01 1914-04-01 ... 1947-07-01


months(d)
[1] "Januar" "März" "September" "Dezember" "Juli"

format(d,"%B") # months alternative
[1] "Januar" "März" "September" "Dezember" "Juli"

format(d,"%Y") # years
[1] "1937" "1916" "1913" "1927" "1947"

Legend variations

par(mar=c(5.1,4.1,4.1,11.1))
plot( x=1:5, y=1:5, type="n", xlab="x", ylab="y" )

legend( x=2, y=6, legend=c("A","B","C")
  , fill=c("red","blue","green")
  , density=30, bty="n", horiz=TRUE
  , xpd=TRUE )

legend( x=2, y=2, xjust=0.5, yjust=0
  , title=" My title:", title.col="grey40", title.adj=0
  , legend=c("A","B","C","D","E")
  , pch=c(22,22,22,45,45), pt.cex=c(1.2,1.2,1.2,2,2)
  , col=c(rep("black",3),"orange","red")
  , pt.bg=c("blue","green","yellow")
  , bg="grey95", cex=0.8
  , box.col="darkgrey", box.lwd=3, box.lty="dotted" )

legend("topright", inset=0.05, cex=0.8, bg="white"
  , legend=c("A-1","A-2","B-1", "B-2")
  , col=c("lightblue","blue","salmon","red"), pch=15, pt.cex=1.5
  , y.intersp=1.5, x.intersp=1.5 , ncol=2 )

windowsFonts("sans2"="Arial Black")
usr <- par(font=4, family="sans2" )
legend( x=5.5, y=3, legend=c("Label A","Label B","Label C")
  , fill=c("red","orange","yellow")
  , border="brown"
  , y.intersp=2, text.width=strwidth("Make larger")
  , text.col=c("red","orange","yellow")
  , xpd=TRUE )
par(usr)

legend( x="bottomleft", inset=0.02, legend=c("A","B","C","D")
  , lty=c("dashed","dotted",NA,"solid"), lwd=2, cex=0.8
  , pch=c(NA,NA,21,15)
  , col=c("red","blue","black","grey"), bg="white" )

merge

A simple merge example:

a <- data.frame(
  "id"=c(1,2,3,6,7,8),
  "name"=c("Anna","Berta","Claudia","Dora","Eliane","Frida"))
b <- data.frame(
  "id"=c(1,2,3,4,5,9),
  "ort"=c("Zürich","Davos","Zermatt","Chamonix","Verbier","Sedrun"))
c.inner <- merge(a,b)
c.left <- merge(a,b, all.x=TRUE)
c.right <- merge(a,b, all.y=TRUE)
c.full <-  merge(a,b, all.x=TRUE, all.y=TRUE)

Split - Apply - Combine

Another effort to make groupwise operations with a split - apply - combine solution (maybe we'll need it someday):

d.frm <- data.frame(
    name=c("Max","Max","Max","Max","Max","Moritz","Moritz","Moritz")
  , typ=c("rot","blau","grün","blau","grün","rot","rot","blau")
  , anz=c(5,4,5,8,3,2,9,1) )

# Split to list
groups <- split(d.frm, list(d.frm$name, d.frm$typ))

# Create result vector
results <- vector("list", length(groups))
# Apply
for(i in seq_along(groups)) {
  groups[[i]] <- transform(groups[[i]],
    rank = rank(-anz, ties.method = "first"))
  results[[i]] <- groups[[i]]
}

# Combine
result <- do.call("rbind", results)
result
     name  typ anz rank
2     Max blau   4    2
4     Max blau   8    1
8  Moritz blau   1    1
3     Max grün   5    1
5     Max grün   3    2
6     Max  rot   5    1
61 Moritz  rot   2    2
7  Moritz  rot   9    1

We might want to try here the elegant function ave as well:

d.frm$rank_g <- ave( -d.frm$anz, d.frm$name, d.frm$typ,
  FUN=function(x) rank(x, ties.method="first") )

... and not to forget the specially R-flavour:

split(x, g) <- lapply(split(x, g), FUN) 

Dotchart nach Cleveland

Alternativ zum Barplot, wenn Mittelwerte oder Mediane gruppenweise dargestellt werden sollen:

dotchart( tapply( ToothGrowth$len,  list(ToothGrowth$supp, ToothGrowth$dose), mean )
  , main="Guinea Pigs' Tooth Growth", cex=0.8
  , xlab="Vitamin C dose", ylab="tooth length")

Stripchart

Als Alternative zum Boxplot mit wenigen Datenpunkten eignet sich der Stripchart.

stripchart( len ~ dose, data=ToothGrowth, method="jitter", vertical=T
  , subset = supp == "VC", col="blue", main="Guinea Pigs' Tooth Growth"
  , xlab="Vitamin C dose", ylab="tooth length")

stripchart( len ~ dose, data=ToothGrowth, method="jitter", vertical=T
  , subset = supp == "OJ", col="red", add=T)

Grösse des Plot-Fensters bestimmen

Die Grösse des Plot-Fensters kann mit windows() festgelegt werden. Das Grafikfenster wird mit dev.off() wieder geschlossen.

windows(width=10, height=5)
plot( 1:5 )
dev.off()

Selektieren von Listenelementen

Listenelemente können systematisch mit lapply( list, "[", element ) extrahiert werden.

d.frm <- data.frame( id_name=c("1-Max","2-Maria","3-Steven","4-Jane"))

# split into list
lst <- strsplit( x=as.character(d.frm$id_name), split="-" )
# get first elements out of list
d.frm$id <- as.integer( unlist( lapply( lst, "[", 1) ))
d.frm$name <- unlist( lapply( lst, "[", 2) )

d.frm
   id_name   name id
1    1-Max    Max  1
2  2-Maria  Maria  2
3 3-Steven Steven  3
4   4-Jane   Jane  4

str(d.frm)
'data.frame':   4 obs. of  3 variables:
 $ id_name: Factor w/ 4 levels "1-Max","2-Maria",..: 1 2 3 4
 $ name   : chr  "Max" "Maria" "Steven" "Jane"
 $ id     : int  1 2 3 4

Reduce margin between plot region and axes with xaxs, yaxs

# get some data
x.i <- seq(0,1,length=5); y.i <- c( 0,0.1,0.2,0.8,1)
par(mfrow=c(1,2))

plot( y=y.i, x=x.i, type="s", panel.before=grid())
symbols( x=0, y=0, circles=0.12, inches=F, add=T, xpd=T, bg=rgb(0,0,1,0.2) )

plot( y=y.i, x=x.i, type="s", xaxs="i", yaxs="i")
grid(); box()
symbols( x=0, y=0, circles=0.12, inches=F, add=T, xpd=T, bg=rgb(0,0,1,0.2) )

Find most frequent elements

# the vector
x <- sample.int( n=10, size=20, replace=TRUE )
# the 3 most frequent elements
names( head( sort(-table(x)), 3 ) )
# the 3 most frequent elements with their frequencies
head( sort(-table(x)), 3 )

Groupwise boxplot

Groupwise boxplots can easily be created by means of the formula interface.

boxplot(len ~ supp*dose, data = ToothGrowth,
        main = "Guinea Pigs' Tooth Growth",
        xlab = "Vitamin C dose mg", ylab = "tooth length",
        col=c("yellow", "orange") 
        )

Why an outdated method is described in the boxplot help is however not directly clear. Maybe we are glad to know about the technique anyway someday...

boxplot(len ~ dose, data = ToothGrowth,
       boxwex = 0.25, at = 1:3 - 0.15,
       subset = supp == "VC", col = "yellow",
       main = "Guinea Pigs' Tooth Growth",
       xlab = "Vitamin C dose mg",
       ylab = "tooth length",
       xlim = c(0.5, 3.5), ylim = c(0, 35), yaxs = "i")

boxplot(len ~ dose, data = ToothGrowth, add = TRUE,
       boxwex = 0.25, at = 1:3 + 0.15,
       subset = supp == "OJ", col = "orange")

legend(2, 9, c("Ascorbic acid", "Orange juice"),
      fill = c("yellow", "orange"))

SQL-OLAP in R

How to generate SQL-OLAP functions in R:

d.frm <- data.frame( x=rep(1:4,3), g=gl(4,3,labels=letters[1:4]) )

# SQL-OLAP: sum() over (partition by g)
# (more than 1 grouping variables are enumerated like ave(..., g1,g2,g3, FUN=...)):
d.frm$sum_g <- ave( d.frm$x, d.frm$g, FUN=sum )


# same with rank (decreasing):
d.frm$rank_g <- ave( -d.frm$x, d.frm$g, FUN=rank )
d.frm


# get some more data
d.frm <- data.frame(
  id=c("p1","p1","p2","p2","p2","p3","p2","p3","p1","p1","p2"),
  A=c(0,1,1,1,0,0,0,0,0,0,0),
  B=c(1,0,0,0,0,0,0,0,0,0,0),
  C=c(0,0,0,0,1,1,1,0,1,1,1)
)

# get rownumber by group, based by original order
d.frm$rownr <- ave( 1:nrow(d.frm), d.frm$id, FUN=order )

# get some groupwise aggregation on more than one column 
d.frmby <- data.frame( lapply( d.frm[,-c(1,5)], tapply, d.frm$id, "max", na.rm=TRUE ))

# (see also 'Split - Apply - Combine' post)

Barplot mit Fehlerbalken

hh <- t(VADeaths)[, 5:1]
ci.l <- hh * 0.85
ci.u <- hh * 1.15


mb <- barplot(hh, beside = TRUE, ylim = c(0, 100)
  , col = c("lightblue", "mistyrose","lightcyan", "lavender")
  , main = "Death Rates in Virginia", font.main = 4
  , sub = "Faked 95 percent error bars", col.sub = "gray20"
  , cex.names = 1.5
  , legend.text = colnames(VADeaths), args.legend = list( bg="white" )
  , panel.before = {
      rect( xleft=par()$usr[1], ybottom=par()$usr[3], xright=par()$usr[2], ytop=par()$usr[4]
        , col="gray99" )
      grid( nx=NA, ny=NULL ) # horiz grid only
      box()
  }
  , xpd=F )


arrows( x0=mb, y0=ci.l, y1 = ci.u, angle=90, code=3, length=0.05 )


mtext( side = 1, at = colMeans(mb), line = 2,
text = paste("Mean", formatC(colMeans(hh))), col = "red" )

Simple textplot

Putting text on a plot is not that straight on, especially if there's more than one line of text.

data(iris)

# put the summary output into a variable
out <- capture.output( 
  summary(lm(Sepal.Length ~ Species + Petal.Width, iris)) )

cat( out, sep="\n" )

# create plot
plot.new()

# print text in mono font
text(labels=out, x=0
  , y=rev(1:length(out)) * strheight( "S", cex=0.8 ) * 1.3
  , adj=c(0,0), family="mono", cex=0.8 )



This is what I thought so far. Of course it is straight on to place bulk text on several lines, stupid...
Just collapse the text with newline as separator:

text(labels=paste(out, collapse="\n"), x=0, y=0
      , adj=c(0,0), family="mono", cex=0.8 )

Format

Tausender-Trennzeichen und Nachkommastellen
format( 1234, big.mark="'", nsmall=2 )
[1] "1'234.00"


leading zeros
sprintf("%04d", 15)
[1] "0015"

oder mit formatC:
formatC(1:5, width=2, flag="0")
[1] "01" "02" "03" "04" "05"


date
format( Sys.time(), "%Y-%m-%d %H:%M:%S" )
[1] "2010-05-19 10:31:57"

numers as hexnumbers
sprintf("%1$d %1$x %1$X", 0:255)

Codes unter R-Help: strptime 

Von R zu Word

Control MS Word from R, do some reporting. The package RDCOMClient is great for that.
The library DescTools (available on CRAN) uses this package and contains some wrapping functions to make things easier.

List in DataFrame

d.frm <- data.frame(name=c("Max","Max","Max","Max","Max","Moritz","Moritz","Moritz")
,typ=c("rot","blau","grün","blau","grün","rot","rot","blau")
,anz=c(5,4,5,8,3,2,9,1) )

d.frm <- d.frm[ order(d.frm$name, -d.frm$anz),]
d.frm


Zusammenfassung wie summaryBy:

> as.data.frame( do.call("rbind", tapply( d.frm$anz, d.frm$name, summary )))

       Min. 1st Qu. Median Mean 3rd Qu. Max.
Max       3     4.0      5    5    5.0     8
Moritz    1     1.5      2    4    5.5     9

Windows Schriftarten verwenden

Windows Schriftarten müssen definiert werden, bevor sie genutzt werden können:

# Type family examples - creating new mappings
plot(1:10,1:10,type="n")
windowsFonts(
A=windowsFont("Arial Black"),
B=windowsFont("Bookman Old Style"),
C=windowsFont("Comic Sans MS"),
D=windowsFont("Symbol")
)

text(3,3,"Hello World Default")
text(4,4,family="A","Hello World from Arial Black")
text(5,5,family="B","Hello World from Bookman Old Style")
text(6,6,family="C","Hello World from Comic Sans MS")
text(7,7,family="D", "Hello World from Symbol")

Add a time axis

Adding a time scaled x-axis to a plot can be tedious...

# add a time axis
axis(side = 1, at  =as.numeric(seq(as.Date("2000-01-01"), by = "year", length = 10))
  , labels = format( seq(as.Date("2000-01-01"), by = "year", length = 10) , "%Y")
  , cex.axis = 0.8, las = 1)

# plot the timegrid
# vertical gridlines according to chosen time-values
abline(v = seq(as.Date("2000-01-01"), by = "year", length = 10), col = "grey", lty = "dotted")
# where to plot the horizontal gridlines? get the coords by means of axTicks()
abline(h = axTicks(2), col = "grey", lty = "dotted")

Dummy-Kodierung

Gesucht ist die Dummy-Kodierung eines Faktors:

d.frm <- data.frame(
    name=c("Max","Max","Max","Max","Max","Moritz","Moritz","Moritz")
  , typ=c("rot","blau","grün","blau","grün","rot","rot","blau")
  , anz=c(5,4,5,8,3,2,9,1)
)

d.frm
    name  typ anz
1    Max  rot   5
2    Max blau   4
3    Max grün   5
4    Max blau   8
5    Max grün   3
6 Moritz  rot   2
7 Moritz  rot   9
8 Moritz blau   1

Gewünschtes liefert

model.matrix(~d.frm$typ)[,-1]
  d.frm$typgrün d.frm$typrot
1             0            1
2             0            0
3             1            0
4             0            0
5             1            0
6             0            1
7             0            1
8             0            0

oder alternativ die Funktion class.ind aus der library(nnet):

library(nnet)
class.ind( df$typ )

blau grün rot
[1,] 0 0 1
[2,] 1 0 0
[3,] 0 1 0
[4,] 1 0 0
[5,] 0 1 0
[6,] 0 0 1
[7,] 0 0 1
[8,] 1 0 0

Still another brilliant Ripley solution:

ff <- factor(sample(letters[1:5], 25, replace=TRUE))
diag(nlevels(ff))[ff,]

Read from/Write to clipboard

Daten lesen aus dem, resp. schreiben in das Clipboard ganz einfach:

d.x <- data.frame( x=c(1,2,3,3), y=c("A","B","A","A") )
write.table(d.x, file="clipboard", sep="\t", row.names=FALSE)
a <- read.table("clipboard", header=TRUE)

Modell-Formelnotation

---------------------------------------------------------------------------------------------
Operator Bedeutung
----------------------------------------------------------------------------------------------
+... Hinzunahme einer Variablen
-... Herausnahme einer Variablen (-1 für Achsenabschnitt)
:... Wechselwirkung/Interaktion von Variablen
*... Hinzunahme von Variablen und deren Wechselwirkungen
/... hierarchisch untergeordnet ("nested").
.... y/z bedeutet: z hat nur Wirkung innerhalb der Stufen von y, aber nicht global.
^... alle Interaktionen bis zum angegebenen Grad
.... alle Variablen aus dem Datensatz in das Modell aufnehmen
I(). innerhalb von I() behalten arithmetische Operatoren ihre ursprüngliche Bedeutung
.... ("Inhibit interpretation")
----------------------------------------------------------------------------------------------

Bsp:
y ~ x1 * x2 * x3 - x1:x2:x3 # ohne 3-fach Interaktion
y ~ (x1 + x2 + x3)^2 # alle Variablen inkl. Interaktion bis zum 2. Grad

Kombinationen tabellieren

Ausgangslage ist ein data.frame mit einer Zeile pro Element und n Indikatorvariablen. Es stellt sich die Frage nach den Häufigkeiten der Kombinationen der Indikatorvariablen.

d.frm <- data.frame(name=c("Franz", "Maria", "Claudine")
  , A=c(1,0,1), B=c(1,1,1), C=c(0,0,1), D=c(0,1,0))

Solution 1 (clumsy):

# Berechne die Summe der mit einer 2er-Potenz multiplizierten Indikatoren
d.frm$sum <- as.matrix(d.frm[,2:5]) %*% (2^(0:3))

# Erzeuge levels für eine Komb-Faktor
d.lvl <- expand.grid( c(0,"A"),c(0,"B"),c(0,"C"),c(0,"D") )

# Bilde eine neue Variable mit der Textbezeichnung der 2^n-Summe
d.frm$sum_x <- factor(d.frm$sum
  , levels=((d.lvl!=0)*1) %*% (2^(0:3))
  , labels=gsub( "0","", do.call( "paste", c( d.lvl, list(sep="") ))))

d.frm
......name A B C D sum sum_x
1 ...Franz 1 1 0 0 ..3 ...AB
2 ...Maria 0 1 0 1 .10 ...BD
3 Claudine 1 1 1 0 ..7 ..ABC

Solution 2 (smart):

d.frm$sum_xx <- apply( d.frm[,2:5], 1,
  function(x) paste(LETTERS[1:4][as.logical(x)], collapse="") )


Wenn einfach nur die Kombinationen gesucht sind, gibt's dafür die Bordmittel:

> combn( letters[1:4], 2 )
[,1] [,2] [,3] [,4] [,5] [,6]
[1,] "a" "a" "a" "b" "b" "c"
[2,] "b" "c" "d" "c" "d" "d"

Oder auch interessant mit den Funktionen outer und lower.tri (pairwise):

m <- outer(x, x, paste, sep="-" )
m[!lower.tri(m, diag=TRUE) ]

[1] "a-b" "a-c" "b-c" "a-d" "b-d" "c-d"

Get all binary combinations
The idea is to get a vector with n 0s and n 1s, chop it into n parts of c(0,1) (this is a list), and use expand.grid:

n <- 4="4">
expand.grid(split(rep(c(0,1), each=n), 1:n))

Gruppenweise Auswertung

Erstaunlich umständlich ist die gruppenweise Selektion von bestimmten Elementen eines dataframes. Zum Beispiel soll gruppenweise das erste Element nach einer vorgegebenen Sortierung zurückgegeben werden:

d.frm <- data.frame(
name=c("Max","Max","Max","Max","Max","Moritz","Moritz","Moritz")
,typ=c("rot","blau","grün","blau","grün","rot","rot","blau")
,anz=c(5,4,5,8,3,2,9,1) )

d.frm <- d.frm[ order(d.frm$name, -d.frm$anz),]
d.frm
....name .typ anz
4 ...Max blau ..8
1 ...Max .rot ..5
3 ...Max grün ..5
2 ...Max blau ..4
5 ...Max grün ..3
7 Moritz .rot ..9
6 Moritz .rot ..2
8 Moritz blau ..1


d.frm[ tapply( rownames(d.frm), d.frm$name, head, n=1), ]
. ..name. typ.anz
4....Max.blau...8
7 Moritz..rot...9

Noch einfacher ist allerdings:
d.frm <- d.frm[ order(d.frm$name, -d.frm$anz),]
d.frm[ !duplicated(d.frm$name),]

Wenn der zweite Wert gesucht wäre, ginge dies mit der Index-Funktion "[":
d.frm[ tapply( rownames(d.frm), d.frm$name, "[", 2),  ]

Umorganisation von Daten

Daten, die pivotisiert vorliegen, können mit reshape (siehe auch stack/unstack) wieder in die Faktor-Form gebracht werden.

# Ausgangslage: Breite Form
> d.wide <- data.frame( age=c(22,34,28,31), gender_c=c("m","w","w","m"), ZH=c(12,33,2,5), BE=c(5,6,3,1))

> d.wide

.age gender_c ZH BE
1 22 ....m 12 .5
2 34 ....w 33 .6
3 28 ....w .2 .3
4 31 ....m .5 .1

# ...und die lange Form:

> d.long <- reshape( d.wide, varying=3:4, times=names(d.tmp)[-c(1:2)], v.names="count", direction="long")

> d.long
age gender_c time count id
1.ZH 22 m ZH 12 1
2.ZH 34 w ZH 33 2
3.ZH 28 w ZH 2 3
4.ZH 31 m ZH 5 4
1.BE 22 m BE 5 1
2.BE 34 w BE 6 2
3.BE 28 w BE 3 3
4.BE 31 m BE 1 4

... und wieder zurück:
wide <- reshape( d.long, idvar=c("age","gender_c","id"), timevar="time", direction="wide" )
wide

Ränder-Problematik

Einstellungen für Ränder in Grafiken für: mar, mai, mgp, omi, oma.
mar=c(5,4,4,2)+0.1 gleich mar=c(u,li,o,re)+0.1,

mai=c(u,li,o,re) Angabe in inch,

mgp=c(3,1,0) gleich c(Titel, Label, Achse) in mex-Einheiten,

omi=c(0,0,0,0) gleich omi=c(u,li,o,re),

oma=c(0,0,0,0) gleich oma=c(u,li,o,re) in Textzeileneinheiten.

Quelle: http://www.geo.fu-berlin.de/geol/fachrichtungen/pal/mitarbeiter/plank/formeln_in_R.pdf

Gauss-Test

Beispiel
Das Gewicht von Briefumschlägen für Luftpost kann als normalverteilt angesehen werden, wobei das mittlere Gewicht weniger als 2 g betragen soll. Die Standardabweichung ist erfahrungsgemäss s=0.02 g. Ein Hersteller entnimmt der Produktion 100 Umschläge und stellt den Mittelwert mw_sp=1.98 g fest.

# Gausstest
n <- 100 # Anzahl Umschläge
MW <- 2 # Mittelwert der Grundgesamtheit
SD <- 0.02 # Standardabweichung der Grundgesamtheit
mw <- 1.98 # Mittelwert der Stichprobe
se <- SD/sqrt(n) # Standardfehler des Mittelwerts
alpha <- 0.05 # Signifikanzniveau

# Quantile der normalverteilung
qnorm(alpha/2, mean=MW, sd=se)
# p-Wert des zweiseitigen Gauss-Testes
2*min(pnorm(mw, mean=MW, sd=se), 1-pnorm(mw, mean=MW, sd=se))