Exact Tests memory problem
Andreas Voelp
Since the availability of the Exact Tests option, SPSS keeps reporting memory
problems every once in a while - with data set sizes that should really work!
For example, I'm trying to compute an exact Mann-Whitney test on a sample of
approximately 2 x 35 observations. SPSS gives me only the asymptotic test and
the following message: 'Some or all exact significances cannot be computed
because there is insufficient memory.'
I'm currently using SPSS 8.0.2 running under Win 95 OSR2 on a Pentium-II with
128 MB RAM. I've been experimenting with various combinations of SPSS' MXMEMORY
and WORKSPACE settungs. MXMEMORY reports 14,336K bytes, which should *really* be
enough for that problem.
I first started usimg Exact Tests with version 6.x (I don't remember the exact
version when the option was initially released) under Windows 3.11. The problem
has been accompanying me since that version, through 7.x, until 8.x under Win95.
Wait for this: when I run StatXact *before* starting SPSS, SPSS Exact Test runs
*without any problems*!
Any suggestions or ideas?
Regards,
Andreas Voelp.
--
psy consult - Dr. Andreas Voelp
Fuchstanzstrasse 107 - 60489 Frankfurt, Germany
Phone (069) 97840133 - Fax (069) 97840134
eMail PsyConsu...@t-online.de
assets.wharton.upenn.edu
Q: What does it mean if I get an "out of memory" error in
SPSS?
A: SPSS is not really out of memory. Instead, SPSS has run out of disk
space! By default, SPSS uses the /tmp directory to store it's temporary
files. Normally, the /tmp directory is plenty large enough. However, if
you are running a particularly large job, or if you are using a very large
data file, you may need to tell SPSS to store it's temporary files in a
directory with more space. You can tell SPSS where to store temporary
files by setting the SPSSTEMPDIR variable.
You can change the variable to the /sastemp directory (which is MUCH
larger) by typing:
export SPSSTEMPDIR=/sastemp (this is Unix)
at the Unix prompt _before_ running SPSS. If you are using csh or tcsh
instead of the default Korn shell, enter:
setenv SPSSTEMPDIR /sastemp (this is Unix)
Probably that other program has more disk space and set that up as the
default Temp Disk.
Or in other words, check your head space :)
Andreas Voelp
probably be enough for the size of problem that I'm talking about.
Thanks anyway,
Andreas Voelp
--
Jonathan Fry
>
> Dear list:
>
> Since the availability of the Exact Tests option, SPSS keeps reporting memory
> problems every once in a while - with data set sizes that should really work!
> For example, I'm trying to compute an exact Mann-Whitney test on a sample of
> approximately 2 x 35 observations. SPSS gives me only the asymptotic test and
> the following message: 'Some or all exact significances cannot be computed
> because there is insufficient memory.'
>
> I'm currently using SPSS 8.0.2 running under Win 95 OSR2 on a Pentium-II with
> 128 MB RAM. I've been experimenting with various combinations of SPSS' MXMEMORY
> and WORKSPACE settungs. MXMEMORY reports 14,336K bytes, which should *really* be
> enough for that problem.
This may be part of the problem. Try it with these settings at their
defaults.
>
> I first started usimg Exact Tests with version 6.x (I don't remember the exact
> version when the option was initially released) under Windows 3.11. The problem
> has been accompanying me since that version, through 7.x, until 8.x under Win95.
>
> Wait for this: when I run StatXact *before* starting SPSS, SPSS Exact Test runs
> *without any problems*!
To perform Exact Tests, SPSS runs a piece of StatXact. It might be that
there is not enough memory to load that piece if you don't run StatXact
first, but it's already loaded if you do.
>
> Any suggestions or ideas?
>
> Regards,
> Andreas Voelp.
>
assets.wharton.upenn.edu
: The amount of free disk space on SPSS' temporary device is 1.5 GB - which should
: probably be enough for the size of problem that I'm talking about.
: Thanks anyway,
: Andreas Voelp
Then if that is set for 1.5 GB, the only thing left is the memory allocation.
14 MB's of RAM might be a little shy for an Exact Test, at times. As I
vaguely recall, I ran this on an 8 MB and a 20 MB computer, and it ran or
did not run, depending on the number of variables I requested.
If you have 128 MB of RAM, have you tried running this with, oh, 100 MB of
RAM free instead of the 14 MB you mention? Or can you not do that?
If that other program run first allows SPSS to run, then it may be
organizing the memory better to Win95 which does not clean up memory
allocations as one would expect. If it were me, I would do a clean boot and
have nothing else running, can you do that?
What did SPSS say about this? Can they give you specifications for memory
usage and possible bugs with Win95 or whatever OS you are using?
I have run the Exact Test and run into the memory problem, depending on the
number of variables. It appeared to exponentially need more memory as the
number of variables increased. From what little I remember, this is what
one would expect from the large, harsh demands of the Exact Test.
Andreas Voelp
>
> > I'm currently using SPSS 8.0.2 running under Win 95 OSR2 on a Pentium-II with
> > 128 MB RAM. I've been experimenting with various combinations of SPSS' MXMEMORY
> > and WORKSPACE settungs. MXMEMORY reports 14,336K bytes, which should *really* be
> > enough for that problem.
>
> This may be part of the problem. Try it with these settings at their
> defaults.
>
The manual tells me that 14,000K MXMEM *is* the default. I've been experimenting
with a variety of MXMEM settings between 100K and 100,000K. Within that range,
MXMEM does not appear to have any influence on SPSS' ability to run the exact
test (i. e. my computation problem is so small that it runs even with 100K
MXMEM). It rather appears that the only factor that has an influence on whether
Exact Tests will or will not run is the condition the system is in *before* I
start SPSS. Depending on what I do before entering SPSS, the same syntax and the
same data set will or will not do the job.
There may be several programs that have the an influence here. I usually run so
many different programs throughout the day that it is impossible for me to
determine which if the leaves the system memory in exactly what condition.
However, the only one with which I can reproduce the effect almost
systematically is StatXact.
Regards,
Andreas Voelp