Print

Print


Hi Sean,

To be honest, I'm not quite sure which it is. I do change the parameter
settings because i set seed.0=time but it's definitely possible that the
Java VM might be the problem.

Anyhow, I've started a script similar to the one you suggested. If the issue
presents itself, I will check to see if it's reproducible and send the
details you requested.

Thanks :)

-Mike



On 11/18/05, Sean Luke <[log in to unmask]> wrote:
>
> On Nov 18, 2005, at 6:11 PM, Michael Solano wrote:
>
> > Again, this doesn't occur very often. Most the time, ECJ finds an
> > ideal individual before it quits. I'd say it quits prematurely 1%
> > of the time when running twobox and much less for any of the other
> > problems.
>
> Could you be more specific about this:
> - If you run ECJ with the exact same parameter settings, will it
> fail 1% of the time?
> - or does it only fail 1% of the time when you vary the parameter
> settings?
>
> If it's the first, the problem may *possibly* be the Java virtual
> machine and not ECJ. We've been seeing occasional deaths of the Java
> VM in Linux as well. I'm reasonably confident that ECJ should not
> have different behaviors for different iterations of the same
> parameter settings (assuming seed.0= some val rather than seed.
> 0=time). One strong possibility: running out of memory. We need you
> to output stderr to your current_run.txt as well as stdout.
>
> If it's the second, it could be a bug in ECJ that we'd need to
> stomp. But to do this, we need a specific parameter setting (using a
> specified seed, not seed.0=time) which reproduces the behavior
> consistently. Can you find one? It'd help a lot.
>
> Here's what would help: a script like this:
>
> #! /bin/tcsh
>
> foreach i (0 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 29 30 31 32 33 34 35 36 37 38 \
> 39 40 41 42 43 44 45 46 47 48 49 50)
> echo JOB $i
> java ec.Evolve -file ec/app/twobox/noadf.params \
> -p stat.file = /tmp/stat${i}.txt \
> -p generations=10000000 \
> -p stat.gather-full=true -p seed.0=$i \
> -p stat=ec.simple.SimpleShortStatistics >& /tmp/run${i}.txt
>
> If one of these fails, could you send me the JOB number and the
> /tmp/run<JOB>.txt file that failed?
>
> Sean
>