[R] Shafer's MI software for S-plus

[Prof Brian Ripley]

On Tue, 1 Apr 2003 Ted.Harding at nessie.mcc.ac.uk wrote:

> On 01-Apr-03 Ted Harding wrote:
> > On 01-Apr-03 Prof Brian Ripley wrote:
> >> I have an R port of MIX, but it is not 100% reliable. However, neither
> >> is the S-PLUS original!  (We have found several bugs already.)
> >> 
> >> When I have a few spare days (if ever?) I will try again.  We got
> >> enough to work for the project we needed it for.  Meanwhile,
> >> 
> >> http://www.stats.ox.ac.uk/pub/bdr/mix_1.0-1.tar.gz
> >> 
> >> has the current state.
> > 
> > Thanks, Brian! I'll see how it goes on the somewhat recalcitrant
> > dataset I'm looking at, and report back on anything interesting.
> 
> Got it installed in R (version 1.6.2 on SuSE Linux 7.2). Apparently
> trouble free.
> 
> Worked through the case-study "stlouis.dat" according to Shafer's
> example in the README. All went well until the final batch of
> commands below:
> 
>    # Now try some imputations. The following commands produce three
>    # multiple imputations under the alternative model. The imputations
>    # are proper if we can assume that the data augmentation procedure
>    # achieves stationarity by 100 steps.
>    rngseed(454545)
> ** newtheta<-dabipf.mix(s,margins,design,thetahat4,steps=100,showits=T)
> ** imp1<-imp.mix(s,newtheta,x)
> ** newtheta<-dabipf.mix(s,margins,design,newtheta,steps=100,showits=T)
> ** imp2<-imp.mix(s,newtheta,x)
>    newtheta<-dabipf.mix(s,margins,design,newtheta,steps=100,showits=T)
>    imp3<-imp.mix(s,newtheta,x)
> 
> > newtheta<-dabipf.mix(s,margins,design,newtheta,steps=100,showits=T)
> Steps of Data Augmentation-Bayesian IPF: 
> 1...2...3...4...5...6...7...8...9...10...11...12...13...14...15...
> 16...17...18...19...20...21...22...23...24...25...26...27...28...
> Segmentation fault
> 
> Or it may segfault on the imp1 step or the imp2 step.
> 
> Is this one of the "mix" bugs, or some R/Linux version incompatibility?
> A memory problem (though plating a trace-print of "gc()" indicates that
> there's plenty of R memory to spare)?

It's a bug, but it sometimes does in S-PLUS too.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595