[R] sort (all columns of) a matrix

[Bert Gunter]

Bill:

Defensive programming seems to me to be a wise policy, so thanks for the
helpful tip.

-- Bert

Bert Gunter
Genentech Nonclinical Biostatistics
 
 

-----Original Message-----
From: William Dunlap [mailto:wdunlap at tibco.com] 
Sent: Thursday, October 08, 2009 12:01 PM
To: Bert Gunter; Erik Iverson; Kajan Saied; r-help at r-project.org
Subject: RE: [R] sort (all columns of) a matrix

> -----Original Message-----
> From: r-help-bounces at r-project.org 
> [mailto:r-help-bounces at r-project.org] On Behalf Of Bert Gunter
> Sent: Thursday, October 08, 2009 11:41 AM
> To: 'Erik Iverson'; 'Kajan Saied'; r-help at r-project.org
> Subject: Re: [R] sort (all columns of) a matrix
> 
> Right. My guess is that  Kajan wants:
> 
> a[do.call(order,data.frame(a)),] 
> ## this generalizes to an arbitrary number of columns
> ## do.call() is a very powerful and useful R feature worth 
> learning about

If you are compulsive wrap the second argument to do.call()
with unname(), just in case the data.frame has a column
with a name matching an argument to order (currently
'decreasing' and 'na.last').  This can save you from an
error like the following, where the 'decreasing' column
of 'd' is taken to be the 'decreasing' argument to order,
not just another column to sort:

  > d<-data.frame(weight=c(190, 121, 167, 121),
decreasing=c(TRUE,TRUE,FALSE,FALSE))
  > d[do.call(order,d),]
    weight decreasing
  1    190       TRUE
  3    167      FALSE
  2    121       TRUE
  4    121      FALSE
  > d[do.call(order,unname(d)),]
    weight decreasing
  4    121      FALSE
  2    121       TRUE
  3    167      FALSE
  1    190       TRUE

At some point data.frame may take more pains to make sure its
instances all have column names, in which case the argument needs
to be wrapped with unname(as.list(...)).

(Internally order() can call do.call("order",...) without using unname
so it could run into the same problem.)

Bill Dunlap
Spotfire, TIBCO Software
wdunlap tibco.com 
> 
> 
> Yet another reason why the posting guide asks for a simple, proper,
> reproducible example.
> 
> -- Bert
> 
> Bert Gunter
> Genentech Nonclinical Biostatistics
>