Implement an operator base class for instrumentation

[ascheibe]

Especially for analyzing algorithms and operators, it would be helpful to have operators which can be configured to execute pre- and post processing actions.

[mkommend]

r10149: Implemented instrumentation for operators.

[mkommend]

Added example for operator instrumentation

[mkommend]

Added a sample algorithm which counts the crossover events for a tsp problem. The sample can only be opened when the base class of PermutationCrossover is InstrumentedOperator.

[mkommend]

r10231: Improvements to the InstrumentedOperator class:

• skipped generation of operations if no instrumentation is performed
• changed type of operators parameters to OperatorList

[gkronber]

Discussed in architects meeting.

The instrumentation operations should be execute with a child execution context to enable a parameter lookup for parameters defined in the instrumented operator.

The base class of the following classes should be changed to the InstrumentedOperator:

• SolutionCreators
• Crossovers
• Manipulators
• Evaluators
• Selectors

[mkommend]

r10261: Chained execution contexts of before and after operators in the instrumented operator to enable a parameter lookup.

[mkommend]

r10291: Added interface for instrumented operators and adapted problem and encoding specific operators to provide instrumentation capabilities.

All SolutionCreators, Crossovers, Manipulators, Evaluators and Selecters except the following are now derived from the InstrumentedOperator:

• VRP related operators
• OKB relatored operators
• Move related operators
• UserDefinedOperators
• RealVectorParticleCreator (Encodings.RealVector)
• ExternalEvaluator
• GenderSpecificSelector
• OffspringSelector

[mkommend]

r10292: Reverted accidental changes in OKB.EmptySolutionCreator and UserDefinedProblem.

[mkommend]

Please adapt the VRP related operators.

[mkommend]

r10295: Adapted all MultiOperators to subclass InstrumentedOperator.

[svonolfe]

r10298 : Adapted all VRP related operators to subclass InstrumentedOperator

[svonolfe]

r10299: Adapted all FLA related operators to subclass InstrumentedOperator

[jkarder]

r10309: Adapted MultiIntValueCrossover to subclass InstrumentedOperator.

[gkronber]

I think these two changes are not correct. It seems the selection is already done correctly in the base class (StochasticMultiOperator). Please recheck these changes and correct if necessary:

• ​https://dev.heuristiclab.com/trac/hl/core/changeset/10298/trunk/sources/HeuristicLab.Problems.VehicleRouting/3.4/Encodings/General/Manipulators/BiasedMultiVRPSolutionManipulator.cs
• ​https://dev.heuristiclab.com/trac/hl/core/changeset/10298/trunk/sources/HeuristicLab.Problems.VehicleRouting/3.4/Encodings/General/Crossovers/BiasedMultiVRPSolutionCrossover.cs

Reviewed: r10309,r10399,r10298, r10295, r10292, r10291, r10261, r10231, r10149. Important part is r10149.

In reviewing I noticed that weighted (proportional) sampling is implemented in each StochasticMultiOperator. I think we should use the static methods from HeuristicLab.Random instead. The code looks the same so hopefully there would not be a change in behaviour.

[mkommend]

Replying to gkronber:

I think these two changes are not correct. It seems the selection is already done correctly in the base class (StochasticMultiOperator). Please recheck these changes and correct if necessary:

• ​https://dev.heuristiclab.com/trac/hl/core/changeset/10298/trunk/sources/HeuristicLab.Problems.VehicleRouting/3.4/Encodings/General/Manipulators/BiasedMultiVRPSolutionManipulator.cs
• ​https://dev.heuristiclab.com/trac/hl/core/changeset/10298/trunk/sources/HeuristicLab.Problems.VehicleRouting/3.4/Encodings/General/Crossovers/BiasedMultiVRPSolutionCrossover.cs

These two code duplications are necessary, because the biased operators update the selection probabilities during the algorithm run and therefore could not reuse the already implemented functionality. SVonolfe added a comment in the classes with r10472.

[gkronber]

r10474 using RandomEnumerable.SampleProportional in BiasedMultiVRPOperators. This should also be changed in StochasticMultiOperator but is not easily possible because of plugin dependency HeuristicLab.Random -> HeuristicLab.Operators.

[gkronber]

If there are no further problems this ticket is ready for release.

[gkronber]

Please make sure to merge r10439 into stable again (#2134).

[mkommend]

Replying to gkronber:

r10474 using RandomEnumerable.SampleProportional in BiasedMultiVRPOperators. This should also be changed in StochasticMultiOperator but is not easily possible because of plugin dependency HeuristicLab.Random -> HeuristicLab.Operators.

r10474 is not related to this ticket and will be merged to the stable with #2146.

[mkommend]

r10299: Adapted all FLA related operators to subclass InstrumentedOperator r10309: Adapted MultiIntValueCrossover to subclass InstrumentedOperator.

These two changesets are not applied on the trunk and won't be merged into stable with this ticket.

[mkommend]

r10507: Merged r10149, r10231, r10261, r10291, r10292, r10295 and r10298 into stable.

[abeham]

r10929: merged r10472 to stable (was forgotten in r10507)