Single Random Stratified

Evaluates moving an object to a sample of destination containers rather than to every container. Containers are grouped into similarity classes (scope items), and the sample is drawn evenly from those classes, drastically reducing the destination search space when there are very many containers.

Parameters

Parameter	Type	Required	Default	Description
stratifiedSampleSize	SampleSize	Yes	-	Number of destination containers to sample, distributed evenly across the similarity classes
destinationsToExplore	DestinationsToExploreOptions	No	all scope items	How destination containers are grouped/sampled (e.g. per scope item)
minObjectsToExplore	int	No	1	Minimum number of objects to fully explore before returning the best improving move found so far

Behavior

Given an object to move, the move type samples destination containers from each similarity class (scope item) and evaluates moving the object to the sampled containers, keeping the best move. Sampling a fixed number per class keeps the number of evaluated moves small even when there are tens of thousands of containers.

Once minObjectsToExplore objects have been explored and an improving move has been found, that move is returned; otherwise more objects are explored until one improves the objective or the source container is exhausted. Raising it improves move quality at the cost of roughly minObjectsToExplore * stratifiedSampleSize evaluations per move.

Example

Configure local search to use only the single random stratified move type, sampling 100 destination containers:

SingleRandomStratifiedMoveTypeSpec stratified;

SampleSize sampleSize;
sampleSize.defaultSampleSize() = 100;
stratified.stratifiedSampleSize() = sampleSize;

LocalSearchSolverSpec localSearch;
localSearch.moveTypeList()->push_back(
    ProblemSolver::makeMoveTypeSpec(stratified));

solver.addSolver(localSearch);

(source)