Skip to content

max.col weirdness

4 messages · Daryl Morris, Bert Gunter, Ben Bolker

Original
Daryl Morris
# 
Hi,
I think there's some rounding issue with returning the max column.  
(running 2.9.0 on an Apple, but my buddy found it on his PC)

 > x <- matrix(c(1234.568,1234.569,1234.567),1)
 > max.col(x)
[1] 2
 > x <- matrix(c(12345.568,12345.569,12345.567),1)
 > max.col(x)
[1] 3
 > x <- matrix(c(112345.568,112345.569,112345.567),1)
 > max.col(x)
[1] 3
 > max.col(-x)
[1] 1

Thanks, Daryl
Daryl Morris
# 
Hi,
Sorry, I was too hasty.  I see that the tolerance is in the 
documentation (and that "random" is the default tie-breaker).

Would it be possible to allow tolerance to be a parameter?

thanks, Daryl
Daryl Morris wrote:
Bert Gunter
# 
Try reading the man page, which says:

Details

When ties.method = "random", as per default, ties are broken at random. In
this case, the determination of a tie assumes that the entries are
probabilities: there is a relative tolerance of 1e-5, relative to the
largest (in magnitude, omitting infinity) entry in the row.


Bert Gunter
Genentech Nonclinical Biostatistics

-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On
Behalf Of Daryl Morris
Sent: Thursday, May 28, 2009 1:47 PM
To: r-help at r-project.org
Subject: [R] max.col weirdness

Hi,
I think there's some rounding issue with returning the max column.  
(running 2.9.0 on an Apple, but my buddy found it on his PC)

 > x <- matrix(c(1234.568,1234.569,1234.567),1)
 > max.col(x)
[1] 2
 > x <- matrix(c(12345.568,12345.569,12345.567),1)
 > max.col(x)
[1] 3
 > x <- matrix(c(112345.568,112345.569,112345.567),1)
 > max.col(x)
[1] 3
 > max.col(-x)
[1] 1

Thanks, Daryl

______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
Ben Bolker
#
Daryl Morris wrote:
Could you just use apply(x,1,which.max) ?
[,1]     [,2]     [,3]
[1,] 112345.6 112345.6 112345.6
[1] 2

  It will presumably be less efficient than max.col (which
is implemented in C), and will automatically
pick the first value of the maximum if there are ties, but
should be OK for everyday use?

PS  oddly enough apply(x,1,which.max) is faster -- maybe because
it doesn't try to deal with ties?
test replications user.self sys.self elapsed user.child
2 apply(x, 1, which.max)        50000     7.649    0.456   8.110          0
1             max.col(x)        50000    11.185    0.580  11.828          0
  sys.child
2         0
1         0