No, the current_run.txt does not indicate that an ideal individual has
been found. Usually, it says "Found Ideal Individual" if there is
such an individual.
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.
If I'm not mistaken, the quit-on-run-complete=false parameter that you've added below will cause it to keep going until the maximum number of generations has been reached, regardless of whether or not an ideal individual is found. I'm not sure that will help.
I want the run to stop when it's found a perfect individual (AdjustedFitness=1.0). The only reason I set the max generations so high is to give the run ample time to find a perfect individual.
Do the last lines of the /tmp/current_run.txt file indicate that an
ideal individual has been found? Can you run the following command
and tell me whether it still stops prematurely?
java ec.Evolve -file ec/app/twobox/noadf.params -p stat.file=/tmp/
out.stat -p generations=10000000 -p stat.gather-full=true -p seed.
0=time -p stat=ec.simple.SimpleShortStatistics -p quit-on-run-
complete=false > /tmp/current_run.txt
On Nov 18, 2005, at 4:38 PM, Michael Solano wrote:
> Hi Liviu,
> The out.stat file shows that no individual in the run has achieved
> ideal fitness. The output looks exactly the same as if it had run
> to completion, it just stops at an earlier generation.
> I'm running twobox with pretty much the stock options found in
> noadf.params. I've changed a few parameters through the command
> line as indicated below:
> java ec.Evolve -file ec/app/twobox/noadf.params -p stat.file=/tmp/
> out.stat -p generations=10000000 -p stat.gather-full=true -p seed.
> 0=time -p stat=ec.simple.SimpleShortStatistics > /tmp/current_run.txt
> ECJ Evolutionary Computation Toolkit <[log in to unmask]>
> On 11/18/05, Liviu Panait < [log in to unmask]> wrote:Mike,
> Are you sure it does not find the ideal individual? Can you tell us
> more about the settings you have, and maybe the output you get until
> ECJ crashes?
> On Nov 18, 2005, at 3:59 PM, Michael Solano wrote:
> > Hello,
> > I was wondering if anyone has experienced problems with ECJ
> > prematurely stopping? I run experiments on a cluster of linux
> > computers and every once in a while, I get a run that stops without
> > finding a perfect individual or reaching the maximum number of
> > generations. When this happens, there is no indication an error
> > has occurred. It seems to occur more frequently when running
> > twobox than any of the other included problems.
> > It's possible that the problem has nothing to do with ECJ. It
> > might be due to the cluster I'm using or some modifications I've
> > made to the ECJ code. I haven't been able to narrow it down yet so
> > I wanted to see if anyone else has experienced this problem.
> > Thanks :)
> > -Mike