Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Choosing a target, measure, and algorithm

Start with the question. Choose a target representation and a score, then select a compatible search. The fastest algorithm for an unrelated measure does not answer the intended question.

Target and quality selection

QuestionTargetSuggested starting measureExplanation
Higher/lower frequency of one label?BinaryTargetWRAccQF or StandardQF(a)Binary quality
Higher/lower numeric values?NumericTargetStandardQFNumeric(a)Numeric targets
Different numeric distribution?NumericTargetKolmogorovSmirnovQF, rank or spread measureNumeric targets
Different class composition?MulticlassTargetClassDistributionL1QF or SplitQFMultiple targets
Different numeric mean vector?VectorTargetMeanVectorL1QFMultiple targets
Different marginal distributions of discrete attributes?NominalTargetWeightedKLQFWKL
Different correlation?CorrelationTargetCorrelationDifferenceQFEMM
Different regression slope?RegressionTargetSlopeSignificanceQFEMM
Different multiple-regression coefficients?GeneralRegressionTargetCooksDistanceQFEMM
Different logistic-regression relationship?LogisticRegressionTargetLogisticInteractionQFEMM
Different nominal association?AssociationTargetAssociationDevianceQFEMM
Different conditional output distributions?DecisionTableTargetHellingerQFEMM
Different parameters in a likelihood family?LikelihoodTargetLikelihoodQF(statistic="ratio")EMM
Poorer classifier ranking?SoftClassifierTargetROCAUCQF, PRAUCQF, or ARLQFClassifier auditing
Frequent descriptions without a target?FITargetCountQFSpecialized tasks
Best rule for current loss derivatives?GradientTargetGradientBoostingQFSpecialized tasks

Search selection

SituationStarting choiceGuaranteeImportant qualification
Ordinary target, small/moderate search spaceDFSExhaustiveWithin the configured language
Need alternative exhaustive ordering/pruningOPUS, DpSubgroup, BestFirstSearchExhaustiveCompare on the actual task; memory needs differ
Prefer level-wise candidate generationAprioriExhaustiveLarge frontiers can consume memory
Prefer pattern-tree or vertical organizationSDMapStar, SDMap, VLSDExhaustiveStandard child batching/CUDA is unsupported
Numeric measure with ordering boundsNumBSDExhaustiveRequires a supported numeric quality function
Model recoverable from aggregated statistics, especially EMMGPGrowthExhaustiveCondensed valuations where supported; row sets otherwise
Generalization-aware measureGeneralizationAwareDFS, GeneralizationAwareAprioriExhaustiveThese searches supply the required references
Heuristic search is acceptableBeamSearchHeuristicBeam width does not guarantee the optimum
Diversity should affect exploration and final selectionDSSDHeuristicAlso has different refinement semantics

For the same task, all compatible exhaustive algorithms find the same optimal top-k quality values. Their choice affects runtime, not solution quality; memory requirements and the descriptions selected among equal-quality ties can differ. Heuristic choices can change the attained quality as well as runtime. See the search chapter for the complete algorithm overview.

Exhaustive algorithms with no quality bound can still enumerate, but may be expensive. No general-purpose compatibility table can predict which algorithm will be fastest on every dataset.

Which redundancy or validation operation?

NeedOperation
Remove results covering nearly identical rowsoverlap_filter
Keep one result per set of descriptor featuresunique_attributes
Apply bounds on table statistics after searchstatistic_filter
Favor additions that improve over simpler descriptionsGeneralization-aware quality plus its dedicated search
Favor diversity while exploringDSSD
Test fixed candidates on separate datarandomization_p_values plus adjust_p_values
Split, search, test, and select with a fixed supplied spaceholdout_search

Generalization awareness and diversity do not substitute for validation. A test-based discovery score is also not an automatic correction for the search.