Recent changes to bugs-features

Wrong count of iterations with OptimizationMediator

2013-04-14T05:08:57Z

If the number of iterations of the EA is 1, not a single implementation is evaluated because the "initialize()" methods is considered as iteration. See optimize() method in OptimizationMediator:

public void optimize() throws StopException, TerminationException {
    iterativeOptimizer.initialize();
    nextIteration();
    while (iteration.value() < iteration.max()) {
        iterativeOptimizer.next();
        nextIteration();
    }
}

This method should be:

public void optimize() throws StopException, TerminationException {
    iterativeOptimizer.initialize();
    while (iteration.value() <= iteration.max()) {
        iterativeOptimizer.next();
        nextIteration();
    }
}

But then it needs to be checked that other optimizers also work properly with this change.

#22 Unnamed node in Configurator

2013-02-22T06:38:39Z

status: open --> closed-wont-fix
milestone: -->

#26 Batch.py crashes

2013-02-22T06:36:54Z

status: open --> closed
milestone: -->

Batch.py crashes

2012-10-25T12:53:47Z

If you use multiple processes for result determination, the batch.py script crashes
Command: "python2.7 batch.py -p 8 -s ."

Error:
Traceback (most recent call last):
File "batch.py", line 610, in <module>
if options.results: doResults(basefolder)
File "batch.py", line 451, in doResults
nadir = getMax(nadir)
File "batch.py", line 221, in getMax
return getComp(samples, max)
File "batch.py", line 214, in getComp
sample = Sample(Sample.invValues(minvalues))
File "batch.py", line 103, in invValues
return [v if d == MIN else -v for v, d in zip(values, objectives.dirs)]
AttributeError: 'NoneType' object has no attribute 'dirs'

Batch.py crashes

2012-10-25T12:53:47Z

Ticket 26 has been modified: Batch.py crashes
Edited By: Martin Lukasiewycz (lukasiewycz)
Status updated: u'open' => u'closed'

Dynamically adding objectives leads to AIOFBException

2011-12-21T18:41:29Z

If some objectives need to be feasible for others to be evaluated, the latter have all to be set to infeasible.
Otherwise, the array-based domination check fails:

java.lang.ArrayIndexOutOfBoundsException: 1
at org.opt4j.core.Objectives.dominates(Objectives.java:277)
at org.opt4j.optimizer.ea.Nsga2.fronts(Nsga2.java:205)
at org.opt4j.optimizer.ea.Nsga2.getLames(Nsga2.java:133)
at org.opt4j.optimizer.ea.EvolutionaryAlgorithm.optimize(EvolutionaryAlgorithm.java:151)
at org.opt4j.core.optimizer.Bootstrap.execute(Bootstrap.java:59)
at org.opt4j.start.Opt4JTask.execute(Opt4JTask.java:83)
at org.opt4j.config.Task.call(Task.java:116)
at org.opt4j.config.Task.call(Task.java:1)
at java.util.concurrent.FutureTask$Sync.innerRun(FutureTask.java:303)
at java.util.concurrent.FutureTask.run(FutureTask.java:138)
at java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:886)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:908)
at java.lang.Thread.run(Thread.java:662)

TSVLogger complains about infeasible solutions

2011-10-26T05:37:33Z

See line 187 in TSVLogger: if a value is null, an error message is generated. However, null indicates that the solution is infeasible.

Solution 1:
Delete the error message.
Solution 2:
Ignore the message and do not log the infeasible solutions.

Not adding infeasible solutions to the archive is no solution since this would influence the optimization.

NSGA2 CrowdingDistance Normalization

2011-07-18T09:12:04Z

Normalize the distances for the NSGA2/CrowingDistance (and also CrowingArchive, and Crowding). Pointed out by Qais Noorshams (KIT)

See: K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, "A fast and elitist multiobjective genetic algorithm : Nsga-ii," Evolutionary Computation, IEEETransactions on, vol. 6, no. 2, pp. 182-197, August 2002.

Currently, the code for the calculation of the crowing distance exists twice (due to efficiency reasons of the NSGA2). Investigate whether a single generic CrowdingDistanceAlgorithm.class (currently Crowing.class in org.opt4j.optimizer.mopso package) would be possible without losing too much performance.

Unnamed node in Configurator

2011-07-11T13:32:27Z

If you name an Opt4JModule simply "Module" without any prefix like "MyProblemModule" you get an unnamed node in the configurator.
This is a little GUI flaw.
"Module" should only be stripped iff the resulting node name in the configurator would not be the empty String.

Unnamed node in Configurator

2011-07-11T13:32:27Z

Ticket 22 has been modified: Unnamed node in Configurator
Edited By: Martin Lukasiewycz (lukasiewycz)
Status updated: u'open' => u'closed-wont-fix'