Package | Description |
---|---|
pl.poznan.put.cs.idss.jrs.rules |
Modifier and Type | Class | Description |
---|---|---|
class |
MonotonicVCDomLem |
Class generating decision rules according to VC-DOMLEM algorithm, which is an extension of DOMLEM algorithm,
handling not only DRSA (Dominance-based Rough Set Approach), but also VC-DRSA (Variable Consistency DRSA),
IRSA=CRSA (Indiscernibility-based Rough Set Approach=Classical Rough Set Approach) and VC-IRSA (Variable Consistency IRSA).
This rules generator class is compatible with monotonic single decision classes of type MonotonicDecisionClass and monotonic unions of decision classes of type MonotonicUnion
and uses value of inconsistency measure $\epsilon$ or $\epsilon'$ rule's statistic when searching for best condition added to the induced rule.This class supports: - IRSA + lower approximations, upper approximations or boundaries of decision classes as positive examples (rules' conditions and decisions employ relation '='); this means certain, possible and approximate rules with inconsistency measure value equal to 0, induced for single decision classes - VC-IRSA + lower approximations of decision classes as positive examples (rules' conditions and decisions employ relation '='); this means certain rules with inconsistency measure value possibly greater than 0, induced for single decision classes - DRSA + lower approximations or upper approximations of upward or downward unions of decision classes as positive examples (rules' conditions and decisions employ relations '>=' and '<='); this means certain or possible rules with inconsistency measure value equal to 0, induced for upward or downward unions of decision classes - VC-DRSA + lower approximations of upward or downward unions of decision classes as positive examples (rules' conditions and decisions employ relations '>=' and '<='); this means certain rules with inconsistency measure value possibly greater than 0, induced for upward or downward unions of decision classes TODO - add description of support for DRSA and approximate rules using: 1) mix of conditions from different border objects, 2) conditions taken from two border objects (one object from lower bound and one object from upper bound of considered boundary) - decision table, PCT and similarity table as learning sample (simple values, pairs of simple values and (similarity, reference value) pairs as values of examples' fields for active condition attributes / criteria) - mixed attributes and criteria (attributes with preference type assigned) for DRSA and VC-DRSA - only monotonic VC-DRSA approach, which is described in the article "Monotonic Variable Consistency Rough Set Approaches", Jerzy Błaszczyński, Salvatore Greco, Roman Słowiński, Marcin Szeląg, LNAI 2007, proceedings of the JRS07 conference, Toronto, 14-16 May 2007. - induction based on border (corner) objects (objects from "lower edge" / "upper edge") for VC-DRSA - induction based on all objects for DRSA / IRSA / VC-IRSA - induction of rules using mix of conditions from different objects - induction of robust (object-based) decision rules (method alternative to the one mentioned above) - each induced rule bases entirely on some positive object from learning set (partial profiles of that rule are created on the basis of evaluations from that object; of course not all evaluations from considered object have to be used, but only those, which ensure that rule does not exceed maximal allowed inconsistency measure value) Concluding above, one may give 5 dimensions of the problem: 1) inconsistency measure value = 0 / inconsistency measure value > 0 2) decision classes / upward unions / downward unions 3) certain / possible / approximate rules 4) attributes / criteria 5) SimpleField / PairField / SimilarityField values of examples' fields, for active condition attributes/criteriaFollowing combinations of these 5 dimensions are supported: [IRSA] a) inconsistency measure value = 0; decision classes; certain rules; attributes only; SimpleField values of examples' fieldsb) inconsistency measure value = 0; decision classes; possible rules; attributes only; SimpleField values of examples' fieldsc) inconsistency measure value = 0; decision classes; approximate rules; attributes only; SimpleField values of examples' fields[VC-IRSA] d) inconsistency measure value > 0; decision classes; certain rules; attributes only; SimpleField values of examples' fields[DRSA] g) inconsistency measure value = 0; upward unions; certain rules; criteria only or mixed criteria and attributes; only SimpleField values of examples' fields, only PairField values of examples' fields, only SimilarityField values of examples' fields, or any combination of the former values of examples' fieldsh) inconsistency measure value = 0; upward unions; possible rules; criteria only or mixed criteria and attributes; only SimpleField values of examples' fields, only PairField values of examples' fields, only SimilarityField values of examples' fields, or any combination of the former values of examples' fieldsTODO - add combination for DRSA and approximate rules i) inconsistency measure value = 0; downward unions; certain rules; criteria only or mixed criteria and attributes; only SimpleField values of examples' fields, only PairField values of examples' fields, only PairField values of examples' fields, only SimilarityField values of examples' fields, or any combination of the former values of examples' fieldsj) inconsistency measure value = 0; downward unions; possible rules; criteria only or mixed criteria and attributes; only SimpleField values of examples' fields, only PairField values of examples' fields, only PairField values of examples' fields, only SimilarityField values of examples' fields, or any combination of the former values of examples' fieldsTODO - add combination for DRSA and approximate rules [VC-DRSA] k) inconsistency measure value > 0; upward unions; certain rules; criteria only or mixed criteria and attributes; only SimpleField values of examples' fields, only PairField values of examples' fields, only PairField values of examples' fields, only SimilarityField values of examples' fields, or any combination of the former values of examples' fieldsm) inconsistency measure value > 0; downward unions; certain rules; criteria only or mixed criteria and attributes; only SimpleField values of examples' fields, only PairField values of examples' fields, only PairField values of examples' fields, only SimilarityField values of examples' fields, or any combination of the former values of examples' fieldsPerformance of VC-DOMLEM algorithm is affected by the MonotonicVCDomLem.DEFAULT_MODE_OF_POSITIVE_EXAMPLES_FOR_VCDRSA constant. |
class |
StandardVCDomLem |
Class generating decision rules according to VC-DOMLEM algorithm, which is an extension of DOMLEM algorithm,
handling not only DRSA (Dominance-based Rough Set Approach), but also VC-DRSA (Variable Consistency DRSA),
IRSA=CRSA (Indiscernibility-based Rough Set Approach=Classical Rough Set Approach) and VC-IRSA (Variable Consistency IRSA).
This rules generator class is compatible with standard single decision classes of type StandardDecisionClass and standard unions of decision classes of type StandardUnion
and uses confidence rule's statistic when searching for best condition added to the induced rule.This class supports: - IRSA + lower approximations, upper approximations or boundaries of decision classes as positive examples (rules' conditions and decisions employ relation '='); this means certain, possible and approximate rules with confidence equal to 1, induced for single decision classes - VC-IRSA + lower approximations of decision classes as positive examples (rules' conditions and decisions employ relation '='); this means certain rules with confidence possibly less than 1, induced for single decision classes - DRSA + lower approximations or upper approximations of upward or downward unions of decision classes as positive examples (rules' conditions and decisions employ relations '>=' and '<='); this means certain or possible rules with confidence equal to 1, induced for upward or downward unions of decision classes - VC-DRSA + lower approximations of upward or downward unions of decision classes as positive examples (rules' conditions and decisions employ relations '>=' and '<='); this means certain rules with confidence possibly less than 1, induced for upward or downward unions of decision classes TODO - add description of support for DRSA and approximate rules using: 1) mix of conditions from different border objects, 2) conditions taken from two border objects (one object from lower bound and one object from upper bound of considered boundary) - decision table, PCT and similarity table as learning sample (simple values, pairs of simple values and (similarity, reference value) pairs as values of examples' fields for active condition attributes / criteria) - mixed attributes and criteria (attributes with preference type assigned) for DRSA and VC-DRSA - only new (better) VC-DRSA approach, which is described in the article J. |
class |
VCDomLemOpt |
Optimization of
StandardVCDomLem class. |
class |
VCDomLemOptMV |
VCDomLemOpt class + support for missing values. |