R devel - Feb 2000 - problem with glm (PR#452)

I'm having sporadic problems with glm on my alpha workstation running
Compaq Tru64 UNIX V5.0 (I can't replicate the problem on my Linux PC). I
have the same problem whether I use a version of R built with the native
C-compiler or a version compiled with gcc. In both cases I've built with
g77.

I want to do some simulations with a call to glm in each rep. A simplified
version would be something like:

N_100; NREPS_100; lam.e_5; logit.inv_function(x){ exp(x)/(1+exp(x)) }
for(i in (1:NREPS)) {
   cat(paste("REP",i,"\n"))
   T_rexp(N,(lam.e))
   Delta_rbinom(N,1,logit.inv(1+T))
   ff_glm(Delta ~ T,family=binomial(logit))
}
 
It often completes all reps, but I occasionally get errors related to
lengths of various objects, etc.:

REP 1 
REP 2
<snip>
REP 9 
Error in model.matrix(t, data) : variable lengths differ


REP 1 
<snip>
REP 53 
Error: subscript out of bounds

REP 1 
<snip>
REP 41 
Error in model.matrix(t, data) : invalid variable type

Different error messages, different numbers of reps. When I call glm again
using the values of T and Delta just before the error, it runs error-free.
When I put the data in a data frame before calling glm I can also get
messages like:

REP 1 
<snip>
REP 64 
Error in model.frame(formula, rownames, variables, varnames, extras, extranames,
:
        invalid variable type


--please do not edit the information below--

Version:
 platform = alpha-dec-osf5.0
 arch = alpha
 os = osf5.0
 system = alpha, osf5.0
 status = Patched
 major = 0
 minor = 99.0
 year = 2000
 month = February
 day = 9
 language = R

Search Path:
 .GlobalEnv, Autoloads, package:base

-------------------------------------------------------------------------
Brad McNeney                               email: mcneney@stat.sfu.ca
Department of Mathematics and Statistics   web: www.math.sfu.ca/~mcneney/
Simon Fraser University                    phone: 604.291.4815
8888 University Drive                      fax: 604.291.4947
Burnaby, British Columbia
Canada V5A 1S6


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

mcneney@cs.sfu.ca writes:
> N_100; NREPS_100; lam.e_5; logit.inv_function(x){ exp(x)/(1+exp(x)) }
> for(i in (1:NREPS)) {
>    cat(paste("REP",i,"\n"))
>    T_rexp(N,(lam.e))
>    Delta_rbinom(N,1,logit.inv(1+T))
>    ff_glm(Delta ~ T,family=binomial(logit))
> }
>  
> It often completes all reps, but I occasionally get errors related to
> lengths of various objects, etc.:
...
> --please do not edit the information below--
> 
> Version:
>  platform = alpha-dec-osf5.0
>  arch = alpha
>  os = osf5.0
>  system = alpha, osf5.0
>  status = Patched
>  major = 0
>  minor = 99.0
>  year = 2000
>  month = February
>  day = 9
>  language = R
> 
> Search Path:
>  .GlobalEnv, Autoloads, package:base
I just did the above ~10 times on i386 and nothing happened. We've had
previous reports on trouble in the lm/glm/model.matrix etc. region on
Alpha. 

If you compile with all warnings on, are you getting warnings on
pointer/integer comparisons or assignments?

Can you create a reproducible case (with set.seed to keep the random
generator sequence constant) and would you know how run a debugger on
your version so that we can figure out what is going wrong?

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard@biostat.ku.dk)             FAX: (+45) 35327907
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._