ECJ-INTEREST-L Archives

November 2005

ECJ-INTEREST-L@LISTSERV.GMU.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Sean Luke <[log in to unmask]>
Reply To:
ECJ Evolutionary Computation Toolkit <[log in to unmask]>
Date:
Fri, 18 Nov 2005 19:53:05 -0500
Content-Type:
text/plain
Parts/Attachments:
text/plain (45 lines)
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

ATOM RSS1 RSS2