Ranking within factor subgroups

From: Peter Dalgaard <p.dalgaard at biostat.ku.dk>
To: "maneesh deshpande" <dmaneesh at hotmail.com>
CC: ramasamy at cancer.org.uk, r-help at stat.math.ethz.ch
Subject: Re: [R] Ranking within factor subgroups
Date: 23 Feb 2006 07:28:13 +0100

"maneesh deshpande" <dmaneesh at hotmail.com> writes:

Hi Adai,

I think your solution only works if the rows of the data frame are 

ordered

by "date" and
the ordering function is the same used to order the levels of
factor(df$date) ?
It turns out (as I implied in my question) my data is indeed organized 

in

this manner, so my
current problem is solved.
In the general case, I suppose, one could always order the data frame by
date before proceeding ?

Thanks,

Maneesh

You might prefer to look at split/unsplit/split<-, i.e. the z-scores
by group line:

     z <- unsplit(lapply(split(x, g), scale), g)

with "scale" suitably replaced. Presumably (meaning: I didn't quite
read your code closely enough)

    z <- unsplit(lapply(split(x, g), bucket, 10), g)

could do it.

From: Adaikalavan Ramasamy <ramasamy at cancer.org.uk>
Reply-To: ramasamy at cancer.org.uk
To: maneesh deshpande <dmaneesh at hotmail.com>
CC: r-help at stat.math.ethz.ch
Subject: Re: [R]  Ranking within factor subgroups
Date: Wed, 22 Feb 2006 03:44:45 +0000

It might help to give a simple reproducible example in the future. For
example

 df <- cbind.data.frame( date=rep( 1:5, each=100 ), A=rpois(500, 100),
                         B=rpois(500, 50), C=rpois(500, 30) )

might generate something like

    date   A  B  C
  1    1  93 51 32
  2    1  95 51 30
  3    1 102 59 28
  4    1 105 52 32
  5    1 105 53 26
  6    1  99 59 37
...    . ... .. ..
495    5 100 57 19
496    5  96 47 44
497    5 111 56 35
498    5 105 49 23
499    5 105 61 30
500    5  92 53 32

Here is my proposed solution. Can you double check with your existing
functions to see if they are correct.

   decile.fn <- function(x, nbreaks=10){
     br     <- quantile( x, seq(0, 1, len=nbreaks+1), na.rm=T )
     br[1]  <- -Inf
     return( cut(x, br, labels=F) )
   }

   out <- apply( df[ ,c("A", "B", "C")], 2,
                 function(v) unlist( tapply( v, df$date, decile.fn ) ) 

)

   rownames(out) <- rownames(df)
   out <- cbind(df$date, out)

Regards, Adai



On Tue, 2006-02-21 at 21:44 -0500, maneesh deshpande wrote:

Hi,

I have a dataframe, x of the following form:

Date            Symbol   A    B  C
20041201     ABC      10  12 15
20041201     DEF       9    5   4
...
20050101     ABC         5  3   1
20050101     GHM       12 4    2
....

here A, B,C are properties of a set symbols recorded for a given 

date.

I wante to decile the symbols For each date and property and
create another set of columns "bucketA","bucketB", "bucketC" 

containing

the

decile rank
for each symbol. The following non-vectorized code does what I want,

bucket <- function(data,nBuckets) {
     q <- quantile(data,seq(0,1,len=nBuckets+1),na.rm=T)
     q[1] <- q[1] - 0.1 # need to do this to ensure there are no 

extra

NAs

     cut(data,q,include.lowest=T,labels=F)
}

calcDeciles <- function(x,colNames) {
nBuckets <- 10
dates <- unique(x$Date)
for ( date in dates) {
  iVec <- x$Date == date
  xx <- x[iVec,]
  for (colName in colNames) {
     data <- xx[,colName]
     bColName <- paste("bucket",colName,sep="")
     x[iVec,bColName] <- bucket(data,nBuckets)
  }
}
x
}

x <- calcDeciles(x,c("A","B","C"))


I was wondering if it is possible to vectorize the above function to

make it

more efficient.
I tried,
rlist <- tapply(x$A,x$Date,bucket)
but I am not sure how to assign the contents of "rlist" to their

appropriate

slots in the original
dataframe.

Thanks,

Maneesh