[R] Problems with rlm

[Ed Hagen]

Erin Hodgess wrote:

> Please don't use c as a vector name.  There is a function c and
> that function gets overwritten.

Thanks for that tip.  I was actually using different variable names in 
my script, but simplified them for the posting.  I just tried it using 
the names 'cpavg' and 'havg'.   The first time I called rlm it worked, 
but I got the following errors the second and third time I ran it:

 > out <- rlm(havg ~ cpavg)
Error: cannot allocate vector of size 4178828 Kb
 > out <- rlm(havg ~ cpavg)
Error in model.matrix.default(mt, mf, contrasts) :
         negative length vectors are not allowed

Liaw, Andy wrote:

 > You have not given us nearly enough info.
 > What OS / R version / MASS version (you are using MASS, no?) are you
 > using?

WinXP Pro 2002. Service Pack 1. 512K RAM.  R v. 1.8.1. MASS v. 7.1-11. 
AMD Athlon XP 2600+, 1.91. GHz.

 > Do you have lots of other stuff in the workspace when
 > you tried to run this?

Yes.  A lot of stuff is needed to produce the variables CPRATIO and HEIGHTZ

 > What does gc() tell you?

After running the script, but before running the code in question:

          used (Mb) gc trigger (Mb)
Ncells 720521 19.3    1073225 28.7
Vcells 209563  1.6     786432  6.0

gc() called during and after running the code in question:

 > d2 <- na.omit(data.frame(CPRATIO, HEIGHTZ, FAMILYID))
 > cavg <- tapply(d2$CPRATIO, d2$FAMILYID, mean)
 > havg <- tapply(d2$HEIGHTZ, d2$FAMILYID, mean)
 > gc()
          used (Mb) gc trigger (Mb)
Ncells 720802 19.3    1166886 31.2
Vcells 210286  1.7     786432  6.0
 > out <- rlm(havg ~ cavg)
Error in model.matrix.default(mt, mf, contrasts) :
         cannot allocate vector of length 2146959361
 > gc()
          used (Mb) gc trigger (Mb)
Ncells 720819 19.3    1166886 31.2
Vcells 210334  1.7     786432  6.0

As I noted in my first posting, rlm works fine with CPRATIO and HEIGHTZ, 
which are much larger datasets, if that's any clue (though even CPRATIO 
and HEIGHTZ are still quite small: 85 cases each).  After running the 
main script, I ran

 > out <- rlm(HEIGHTZ ~ CPRATIO)

about 20 consecutive times with no problems.  Yet running rlm with the 
smaller havg and cavg immediately produces errors, and R will usually 
crash after the second or third attempt.

Many thanks for the quick responses,

Ed.

-- 
Edward H. Hagen                  Institute for Theoretical Biology
phone: +49/30 2093-8649             Humboldt-Universität zu Berlin
fax:   +49/30 2093-8801                         Invalidenstraße 43
http://itb.biologie.hu-berlin.de/~hagen      10115 Berlin, Germany