Skip to main content

Single Random Object Stratified

Evaluates moving a sample of objects into a container, rather than every object. Objects are grouped into similarity classes (via a partition), and the sample is drawn from those groups, drastically reducing the search space when there are very many objects.

A sample of objects, grouped by similarity, is evaluated for moving into a container

Parameters​

ParameterTypeRequiredDefaultDescription
objectsToExploreOptionsObjectsToExploreOptionsYes-How objects are grouped/selected for sampling (e.g. by the groups of a partition)
stratifiedSampleSizeSampleSizeYes-Number of objects to sample (spread across the groups)

For sampling to be meaningful, group objects so that members of the same group have similar dimensions.

Behavior​

Given a target container, the move type samples objects from each similarity group and evaluates moving the sampled objects into the container, keeping the best move. Sampling a fixed number per group keeps the number of evaluated moves small even when the total object count is huge.

Example​

Configure local search to use only the single random object stratified move type, sampling 15 objects from the groups of a "group" partition:

// Group similar objects so the sample is drawn from each group.
solver.addPartition("group", objectToGroup);

SingleRandomObjectStratifiedMoveTypeSpec stratified;

GroupList groupList;
groupList.partitionName() = "group";
ObjectsFromGroupsSpec objectsFromGroups;
objectsFromGroups.groupList() = groupList;
ObjectsToExploreOptions objectsToExplore;
objectsToExplore.set_objectsFromGroupsSpec(objectsFromGroups);
stratified.objectsToExploreOptions() = objectsToExplore;

SampleSize sampleSize;
sampleSize.defaultSampleSize() = 15;
stratified.stratifiedSampleSize() = sampleSize;

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

solver.addSolver(localSearch);

(source)