/* * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, write to the Free Software * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ /* * Rule.java * Copyright (C) 2003 Peter A. Flach, Nicolas Lachiche * * Thanks to Amelie Deltour for porting the original C code to Java * and integrating it into Weka. */ package weka.associations.tertius; import weka.core.Instance; import weka.core.Instances; import weka.core.RevisionHandler; import weka.core.RevisionUtils; import java.io.Serializable; import java.text.DecimalFormat; import java.util.ArrayList; import java.util.Comparator; import java.util.Enumeration; /** * Class representing a rule with a body and a head. * * @author Amelie Deltour * @version $Revision: 1.7 $ */ public class Rule implements Serializable, Cloneable, RevisionHandler { /** for serialization */ private static final long serialVersionUID = -7763378359090435505L; /** The body of the rule. */ private Body m_body; /** The head of the rule. */ private Head m_head; /** Can repeat predicates in the rule ? */ private boolean m_repeatPredicate; /** Maximal number of literals in the rule. */ private int m_maxLiterals; /** Can there be negations in the body ? */ private boolean m_negBody; /** Can there be negations in the head ? */ private boolean m_negHead; /** Is this rule a classification rule ? */ private boolean m_classRule; /** Can there be only one literal in the head ? */ private boolean m_singleHead; /** Number of instances in the data this rule deals with. */ private int m_numInstances; /** Set of counter-instances of this rule. */ private ArrayList m_counterInstances; /** Counter for the counter-instances of this rule. */ private int m_counter; /** Confirmation of this rule. */ private double m_confirmation; /** Optimistic estimate of this rule. */ private double m_optimistic; /** * Constructor for a rule when the counter-instances are not stored, * giving all the constraints applied to this rule. * * @param repeatPredicate True if predicates can be repeated. * @param maxLiterals Maximum number of literals. * @param negBody True if negation is allowed in the body. * @param negHead True if negation is allowed in the head. * @param classRule True if the rule is a classification rule. * @param horn True if the rule is a horn clause. */ public Rule(boolean repeatPredicate, int maxLiterals, boolean negBody, boolean negHead, boolean classRule, boolean horn) { m_body = new Body(); m_head = new Head(); m_repeatPredicate = repeatPredicate; m_maxLiterals = maxLiterals; m_negBody = negBody && !horn; m_negHead = negHead && !horn; m_classRule = classRule; m_singleHead = classRule || horn; } /** * Constructor for a rule when the counter-instances are stored, * giving all the constraints applied to this rule. * The counter-instances are initialized to all the instances in the dataset. * * @param instances The dataset. * @param repeatPredicate True if predicates can be repeated. * @param maxLiterals Maximum number of literals. * @param negBody True if negation is allowed in the body. * @param negHead True if negation is allowed in the head. * @param classRule True if the rule is a classification rule. * @param horn True if the rule is a horn clause. */ public Rule(Instances instances, boolean repeatPredicate, int maxLiterals, boolean negBody, boolean negHead, boolean classRule, boolean horn) { m_body = new Body(instances); m_head = new Head(instances); m_repeatPredicate = repeatPredicate; m_maxLiterals = maxLiterals; m_negBody = negBody && !horn; m_negHead = negHead && !horn; m_classRule = classRule; m_singleHead = classRule || horn; m_numInstances = instances.numInstances(); m_counterInstances = new ArrayList(m_numInstances); Enumeration enu = instances.enumerateInstances(); while (enu.hasMoreElements()) { m_counterInstances.add(enu.nextElement()); } } /** * Returns a shallow copy of this rule. * The structured is copied but the literals themselves are not copied. * * @return A copy of this Rule. */ public Object clone() { Object result = null; try { result = super.clone(); /* Clone the body and the head. */ ((Rule) result).m_body = (Body) m_body.clone(); ((Rule) result).m_head = (Head) m_head.clone(); /* Clone the set of counter-instances. */ if (m_counterInstances != null) { ((Rule) result).m_counterInstances = (ArrayList) m_counterInstances.clone(); } } catch (Exception e) { /* An exception is not supposed to happen here. */ e.printStackTrace(); System.exit(0); } return result; } /** * Test if an instance is a counter-instance of this rule. * * @param instance The instance to test. * @return True if the instance is a counter-instance. */ public boolean counterInstance(Instance instance) { return ((m_body.counterInstance(instance) && m_head.counterInstance(instance))); } /** * Update the number of counter-instances of this rule in the dataset. * This method should be used is the rule does not store its counter-instances. * * @param instances The dataset. */ public void upDate(Instances instances) { Enumeration enu = instances.enumerateInstances(); m_numInstances = instances.numInstances(); m_counter = 0; while (enu.hasMoreElements()) { if (this.counterInstance((Instance) enu.nextElement())) { m_counter++; } } m_head.upDate(instances); m_body.upDate(instances); } /** * Get the confirmation value of this rule. * * @return The confirmation. */ public double getConfirmation() { return m_confirmation; } /** * Get the optimistic estimate of the confirmation obtained by refining * this rule. * * @return The optimistic estimate. */ public double getOptimistic() { return m_optimistic; } /* * Get the expected number of counter-instances of this rule, * calculated from the number of instances satisfying the body and * the number of instances satisfying the negation of the head. * * @return The expected number of counter-instances. */ public double getExpectedNumber() { return (double) m_body.getCounterInstancesNumber() * (double) m_head.getCounterInstancesNumber() / (double) m_numInstances; } /** * Get the expected frequency of counter-instances of this rule. * * @return The expected frequency of counter-instances. */ public double getExpectedFrequency() { return getExpectedNumber() / (double) m_numInstances; } /** * Get the observed number of counter-instances of this rule in the dataset. * * @return The observed number of counter-instances. */ public int getObservedNumber() { if (m_counterInstances != null) { return m_counterInstances.size(); } else { return m_counter; } } /** * Get the observed frequency of counter-instances of this rule in the dataset. * * @return The expected frequency of counter-instances. */ public double getObservedFrequency() { return (double) getObservedNumber() / (double) m_numInstances; } /** * Get the rate of True Positive instances of this rule. * * @return The TP-rate. */ public double getTPRate() { int tp = m_body.getCounterInstancesNumber() - getObservedNumber(); int fn = m_numInstances - m_head.getCounterInstancesNumber() - tp; return ((double) tp / (double) (tp + fn)); } /** * Get the rate of False Positive instances of this rule. * * @return The FP-rate. */ public double getFPRate() { int fp = getObservedNumber(); int tn = m_head.getCounterInstancesNumber() - fp; return ((double) fp / (double) (fp + tn)); } /** * Calculate the confirmation of this rule. */ public void calculateConfirmation() { double expected = getExpectedFrequency(); double observed = getObservedFrequency(); if ((expected == 0) || (expected == 1)) { m_confirmation = 0; } else { m_confirmation = (expected - observed) / (Math.sqrt(expected) - expected); } } /** * Calculate the optimistic estimate of this rule. */ public void calculateOptimistic() { int counterInstances = this.getObservedNumber(); int body = m_body.getCounterInstancesNumber(); int notHead = m_head.getCounterInstancesNumber(); int n = m_numInstances; double expectedOptimistic; /* optimistic expected number of counter-instances */ if (counterInstances <= body - notHead) { expectedOptimistic = (double) (notHead * (body - counterInstances)) / (double) (n * n); } else if (counterInstances <= notHead - body) { expectedOptimistic = (double) (body * (notHead - counterInstances)) / (double) (n * n); } else { expectedOptimistic = (double) ((notHead + body - counterInstances) * (notHead + body - counterInstances)) / (double) (4 * n * n); } if ((expectedOptimistic == 0) || (expectedOptimistic == 1)) { m_optimistic = 0; } else { m_optimistic = expectedOptimistic / (Math.sqrt(expectedOptimistic) - expectedOptimistic); } } /** * Test if this rule is empty. * * @return True if it is the empty rule. */ public boolean isEmpty() { return (m_head.isEmpty() && m_body.isEmpty()); } /** * Give the number of literals in this rule. * * @return The number of literals. */ public int numLiterals() { return m_body.numLiterals() + m_head.numLiterals(); } /** * Add a literal to the body of the rule. * * @param newLit The literal to add. * @return The rule obtained by adding the literal, null if the literal can * not be added because of the constraints on the rule. */ private Rule addTermToBody(Literal newLit) { if (!m_negBody && newLit.negative() || (m_classRule && newLit.getPredicate().isClass()) || (newLit instanceof IndividualLiteral && (((IndividualLiteral) newLit).getType() - m_body.getType()) > 1 && (((IndividualLiteral) newLit).getType() - m_head.getType()) > 1)) { return null; } else { Rule result = (Rule) this.clone(); result.m_body.addElement(newLit); /* Update the counter-instances. */ if (m_counterInstances != null) { for (int i = result.m_counterInstances.size() - 1; i >= 0; i--) { Instance current = (Instance) result.m_counterInstances.get(i); if (!result.m_body.canKeep(current, newLit)) { result.m_counterInstances.remove(i); } } } return result; } } /** * Add a literal to the head of the rule. * * @param newLit The literal to add. * @return The rule obtained by adding the literal, null if the literal can * not be added because of the constraints on the rule. */ private Rule addTermToHead(Literal newLit) { if ((!m_negHead && newLit.negative()) || (m_classRule && !newLit.getPredicate().isClass()) || (m_singleHead && !m_head.isEmpty()) || (newLit instanceof IndividualLiteral && ((IndividualLiteral) newLit).getType() != IndividualLiteral.INDIVIDUAL_PROPERTY)) { return null; } else { Rule result = (Rule) this.clone(); result.m_head.addElement(newLit); /* Update counter-instances. */ if (m_counterInstances != null) { for (int i = result.m_counterInstances.size() - 1; i >= 0; i--) { Instance current = (Instance) result.m_counterInstances.get(i); if (!result.m_head.canKeep(current, newLit)) { result.m_counterInstances.remove(i); } } } return result; } } /** * Refine a rule by adding a range of literals of a predicate, either to * the head or to the body of the rule. * * @param pred The predicate to consider. * @param firstIndex The index of the first literal of the predicate to add. * @param lastIndex The index of the last literal of the predicate to add. * @param addTobody True if the literals should be added to the body. * @param addToHead True if the literals should be added to the head. * @return A list of rules obtained by refinement. */ private SimpleLinkedList refine(Predicate pred, int firstIndex, int lastIndex, boolean addToBody, boolean addToHead) { SimpleLinkedList result = new SimpleLinkedList(); Literal currentLit; Literal negation; Rule refinement; for (int i = firstIndex; i < lastIndex; i++) { currentLit = pred.getLiteral(i); if (addToBody) { refinement = addTermToBody(currentLit); if (refinement != null) { result.add(refinement); } } if (addToHead) { refinement = addTermToHead(currentLit); if (refinement != null) { result.add(refinement); } } negation = currentLit.getNegation(); if (negation != null) { if (addToBody) { refinement = addTermToBody(negation); if (refinement != null) { result.add(refinement); } } if (addToHead) { refinement = addTermToHead(negation); if (refinement != null) { result.add(refinement); } } } } return result; } /** * Refine a rule by adding literal from a set of predictes. * * @param predicates The predicates available. * @return The list of the children obtained by refining the rule. */ public SimpleLinkedList refine(ArrayList predicates) { SimpleLinkedList result = new SimpleLinkedList(); Predicate currentPred; boolean addToBody; boolean addToHead; if (this.numLiterals() == m_maxLiterals) { return result; } if (this.isEmpty()) { /* Literals can be added on both sides of the rule. */ for (int i = 0; i < predicates.size(); i++) { currentPred = (Predicate) predicates.get(i); result.addAll(refine(currentPred, 0, currentPred.numLiterals(), true, true)); } } else if (m_body.isEmpty() || m_head.isEmpty()) { /* Literals can be added to the empty side only. */ LiteralSet side; Literal last; if (m_body.isEmpty()) { side = m_head; addToBody = true; addToHead = false; } else { // m_head.isEmpty() side = m_body; addToBody = false; addToHead = true; } last = side.getLastLiteral(); currentPred = last.getPredicate(); if (m_repeatPredicate) { result.addAll(refine(currentPred, currentPred.indexOf(last) + 1, currentPred.numLiterals(), addToBody, addToHead)); } for (int i = predicates.indexOf(currentPred) + 1; i < predicates.size(); i++) { currentPred = (Predicate) predicates.get(i); result.addAll(refine(currentPred, 0, currentPred.numLiterals(), addToBody, addToHead)); } } else { Literal lastLitBody = m_body.getLastLiteral(); Literal lastLitHead = m_head.getLastLiteral(); Predicate lastPredBody = lastLitBody.getPredicate(); Predicate lastPredHead = lastLitHead.getPredicate(); int lastLitBodyIndex = lastPredBody.indexOf(lastLitBody); int lastLitHeadIndex = lastPredHead.indexOf(lastLitHead); int lastPredBodyIndex = predicates.indexOf(lastPredBody); int lastPredHeadIndex = predicates.indexOf(lastPredHead); Predicate inferiorPred; Predicate superiorPred; int inferiorLit; int superiorLit; addToBody = (m_head.numLiterals() == 1 && (lastPredBodyIndex < lastPredHeadIndex || (lastPredBodyIndex == lastPredHeadIndex && lastLitBodyIndex < lastLitHeadIndex))); addToHead = (m_body.numLiterals() == 1 && (lastPredHeadIndex < lastPredBodyIndex || (lastPredHeadIndex == lastPredBodyIndex && lastLitHeadIndex < lastLitBodyIndex))); if (addToBody || addToHead) { /* Add literals in the gap between the body and the head. */ if (addToBody) { inferiorPred = lastPredBody; inferiorLit = lastLitBodyIndex; superiorPred = lastPredHead; superiorLit = lastLitHeadIndex; } else { // addToHead inferiorPred = lastPredHead; inferiorLit = lastLitHeadIndex; superiorPred = lastPredBody; superiorLit = lastLitBodyIndex; } if (predicates.indexOf(inferiorPred) < predicates.indexOf(superiorPred)) { if (m_repeatPredicate) { result.addAll(refine(inferiorPred, inferiorLit + 1, inferiorPred.numLiterals(), addToBody, addToHead)); } for (int j = predicates.indexOf(inferiorPred) + 1; j < predicates.indexOf(superiorPred); j++) { currentPred = (Predicate) predicates.get(j); result.addAll(refine(currentPred, 0, currentPred.numLiterals(), addToBody, addToHead)); } if (m_repeatPredicate) { result.addAll(refine(superiorPred, 0, superiorLit, addToBody, addToHead)); } } else { //((inferiorPred.getIndex() == superiorPred.getIndex()) //&& (inferiorLit < superiorLit)) if (m_repeatPredicate) { result.addAll(refine(inferiorPred, inferiorLit + 1, superiorLit, addToBody, addToHead)); } } } /* Add other literals. */ if (predicates.indexOf(lastPredBody) > predicates.indexOf(lastPredHead)) { superiorPred = lastPredBody; superiorLit = lastPredBody.indexOf(lastLitBody); } else if (predicates.indexOf(lastPredBody) < predicates.indexOf(lastPredHead)) { superiorPred = lastPredHead; superiorLit = lastPredHead.indexOf(lastLitHead); } else { superiorPred = lastPredBody; if (lastLitBodyIndex > lastLitHeadIndex) { superiorLit = lastPredBody.indexOf(lastLitBody); } else { superiorLit = lastPredHead.indexOf(lastLitHead); } } if (m_repeatPredicate) { result.addAll(refine(superiorPred, superiorLit + 1, superiorPred.numLiterals(), true, true)); } for (int j = predicates.indexOf(superiorPred) + 1; j < predicates.size(); j++) { currentPred = (Predicate) predicates.get(j); result.addAll(refine(currentPred, 0, currentPred.numLiterals(), true, true)); } } return result; } /** * Test if this rule subsumes another rule. * * @param otherRule The other rule. * @return True if the other rule is subsumed. */ public boolean subsumes(Rule otherRule) { if (this.numLiterals() > otherRule.numLiterals()) { return false; } return (m_body.isIncludedIn(otherRule) && m_head.isIncludedIn(otherRule)); } /** * Test if this rule and another rule correspond to the same clause. * * @param otherRule The other rule. * @return True if both rules correspond to the same clause. */ public boolean sameClauseAs(Rule otherRule) { return (this.numLiterals() == otherRule.numLiterals() && this.subsumes(otherRule)); } /** * Test if this rule is equivalent to another rule. * * @param otherRule The other rule. * @return True if both rules are equivalent. */ public boolean equivalentTo(Rule otherRule) { return (this.numLiterals() == otherRule.numLiterals() && m_head.negationIncludedIn(otherRule.m_body) && m_body.negationIncludedIn(otherRule.m_head)); } /** * Test if the body of the rule contains a literal. * * @param lit The literal to look for. * @return True if the literal is contained in the body of the rule. */ public boolean bodyContains(Literal lit) { return m_body.contains(lit); } /** * Test if the head of the rule contains a literal. * * @param lit The literal to look for. * @return True if the literal is contained in the head of the rule. */ public boolean headContains(Literal lit) { return m_head.contains(lit); } /** * Test if this rule is over the frequency threshold. * * @param minFrequency The frequency threshold. * @return True if the rule is over the threshold. */ public boolean overFrequencyThreshold(double minFrequency) { return (m_body.overFrequencyThreshold(minFrequency) && m_head.overFrequencyThreshold(minFrequency)); } /** * Test if the body of the rule is true. * * @return True if the body is always satisfied. */ public boolean hasTrueBody() { return (!m_body.isEmpty() && m_body.hasMaxCounterInstances()); } /** * Test if the head of the rule is false. * * @return True if the body is never satisfied. */ public boolean hasFalseHead() { return (!m_head.isEmpty() && m_head.hasMaxCounterInstances()); } /** * Return a String giving the confirmation and optimistic estimate of * this rule. * * @return A String with the values of the rule. */ public String valuesToString() { StringBuffer text = new StringBuffer(); DecimalFormat decimalFormat = new DecimalFormat("0.000000"); text.append(decimalFormat.format(getConfirmation())); text.append(" "); text.append(decimalFormat.format(getObservedFrequency())); return text.toString(); } /** * Return a String giving the TP-rate and FP-rate of * this rule. * * @return A String with the values of the rule. */ public String rocToString() { StringBuffer text = new StringBuffer(); DecimalFormat decimalFormat = new DecimalFormat("0.000000"); text.append(decimalFormat.format(getConfirmation())); text.append(" "); text.append(decimalFormat.format(getTPRate())); text.append(" "); text.append(decimalFormat.format(getFPRate())); return text.toString(); } /** * Retrun a String for this rule. * * @return The String describing this rule. */ public String toString() { StringBuffer text = new StringBuffer(); text.append(m_body.toString()); text.append(" ==> "); text.append(m_head.toString()); return text.toString(); } /** * Comparator used to compare two rules according to their confirmation value. */ public static Comparator confirmationComparator = new Comparator() { public int compare(Object o1, Object o2) { Rule r1 = (Rule) o1; Rule r2 = (Rule) o2; double conf1 = r1.getConfirmation(); double conf2 = r2.getConfirmation(); if (conf1 > conf2) { return -1; } else if (conf1 < conf2) { return 1; } else { return 0; } } }; /** * Comparator used to compare two rules according to their observed number * of counter-instances. */ public static Comparator observedComparator = new Comparator() { public int compare(Object o1, Object o2) { Rule r1 = (Rule) o1; Rule r2 = (Rule) o2; double obs1 = r1.getObservedFrequency(); double obs2 = r2.getObservedFrequency(); if (obs1 < obs2) { return -1; } else if (obs1 > obs2) { return 1; } else { return 0; } } }; /** * Comparator used to compare two rules according to their optimistic estimate. */ public static Comparator optimisticComparator = new Comparator() { public int compare(Object o1, Object o2) { Rule r1 = (Rule) o1; Rule r2 = (Rule) o2; double opt1 = r1.getOptimistic(); double opt2 = r2.getOptimistic(); if (opt1 > opt2) { return -1; } else if (opt1 < opt2) { return 1; } else { return 0; } } }; /** * Comparator used to compare two rules according to their confirmation and * then their observed number of counter-instances. */ public static Comparator confirmationThenObservedComparator = new Comparator() { public int compare(Object o1, Object o2) { int confirmationComparison = confirmationComparator.compare(o1, o2); if (confirmationComparison != 0) { return confirmationComparison; } else { return observedComparator.compare(o1, o2); } } }; /** * Comparator used to compare two rules according to their optimistic estimate * and then their observed number of counter-instances. */ public static Comparator optimisticThenObservedComparator = new Comparator() { public int compare(Object o1, Object o2) { int optimisticComparison = optimisticComparator.compare(o1, o2); if (optimisticComparison != 0) { return optimisticComparison; } else { return observedComparator.compare(o1, o2); } } }; /** * Returns the revision string. * * @return the revision */ public String getRevision() { return RevisionUtils.extract("$Revision: 1.7 $"); } }