Implement RAPGA

[jkarder]

r8313: initial commit

[jkarder]

r8330: initial version

[jkarder]

r8349: added duplication control

[jkarder]

r8354: forgot to add DuplicateSelector

[jkarder]

r8359:

• adjusted event handling
• improved operator graph
• changed default parameter values

[jkarder]

r8377:

• added support for batch offspring creation
• added population size analyzer
• improved operator graph

[jkarder]

r8378: added offspring success analyzer

[jkarder]

r8379: added parameters

[jkarder]

r8385:

• added operator that evaluates expressions
• added selection pressure analyzer
• fixed bug in counting of evaluated solutions

[jkarder]

r8400: removed batch size parameter initialization in main loop

[jkarder]

r8406:

• added similarity calculators
• adjusted event handling
• reformatted code

[jkarder]

r8407:

• fixed event handling
• adjusted types

[jkarder]

r8622:

• fixed some bugs
• minor code improvements

[jkarder]

r8629: added reference to algorithm description

[jkarder]

r8631: removed similarity calculators

[jkarder]

r8642: integrated branch into trunk

[jkarder]

r8643: removed branch RAPGA

[ascheibe]

Thanks for implementing RAPGA. In my opinion this looks very good, I have only 3 reviewing comments:

• As the DuplicatesSelector is a selection operator, it gets discovered by e.g. the GA as a selection operator. Executing a GA with the DuplicatesSelector fails because the SimilarityCalculator is not set. We could fix this by querying the operators collection and setting the SimilarityCalculator if it was not set by the algorithm or problem. This would fix the exception, the question is then if we want such a selection operator because it would mean that we always select duplicates and in the beginning of a GA run it could very well be that no individual is selected. We could switch the two scopes so that the selected scope contains the unique solutions. This would then allow us to have populations in a GA that only contain unique individuals. But of course the population would shrink with algorithm progress and the GA probably can't handle changing population sizes or empty populations. Maybe we should mark it as Non-Discoverable (where the downside is that we can't use this operator anymore for user-defined algorithms) or maybe this isn't a selection operator and we should remove the Selector interface and base class.

• In the DuplicatesSelector you count the number of elements in marks with marks.Count(x => x). This should be marks.Count().

• The Item Name of the SelectionPressureAnalyzer is PopulationSizeAnalyzer, it should be SelectionPressureAnalyzer.

[jkarder]

r8855: applied some of the changes suggested by ascheibe in comment:27:ticket:1247

[jkarder]

Thanks for your review. It is sufficient to remove the Selector interface as all of the discovery is done using this interface. Even though the DuplicateSelector does not need most of the base class' properties, I decided to still inherit from the SingleObjectiveSelector because of the provided functionality.

[jkarder]

r8856: updated RAPGA JSSP sample

[ascheibe]

Thanks very much, looks good.

[jkarder]

Thanks!