Menu
▾
▴
[DL-Learner SVN] SF.net SVN: dl-learner: [715] trunk/src/dl-learner/org/dllearner
|
From: <jen...@us...> - 2008-03-19 08:11:25
|
Revision: 715
http://dl-learner.svn.sourceforge.net/dl-learner/?rev=715&view=rev
Author: jenslehmann
Date: 2008-03-19 01:11:19 -0700 (Wed, 19 Mar 2008)
Log Message:
-----------
redundancy check bug fixes
changes in heuristic
Modified Paths:
--------------
trunk/src/dl-learner/org/dllearner/algorithms/refexamples/ExampleBasedROLearner.java
trunk/src/dl-learner/org/dllearner/algorithms/refexamples/MultiHeuristic.java
trunk/src/dl-learner/org/dllearner/examples/Carcinogenesis.java
trunk/src/dl-learner/org/dllearner/reasoning/FastInstanceChecker.java
trunk/src/dl-learner/org/dllearner/refinementoperators/RhoDRDown.java
Modified: trunk/src/dl-learner/org/dllearner/algorithms/refexamples/ExampleBasedROLearner.java
===================================================================
--- trunk/src/dl-learner/org/dllearner/algorithms/refexamples/ExampleBasedROLearner.java 2008-03-17 10:21:51 UTC (rev 714)
+++ trunk/src/dl-learner/org/dllearner/algorithms/refexamples/ExampleBasedROLearner.java 2008-03-19 08:11:19 UTC (rev 715)
@@ -115,7 +115,7 @@
// the divide&conquer approach in many ILP programs using a
// clause by clause search; after a period of time the candidate
// set is reduced to focus CPU time on the most promising concepts
- private boolean useCandidateReduction = false;
+ private boolean useCandidateReduction = true;
private int candidatePostReductionSize = 30;
// setting to true gracefully stops the algorithm
@@ -244,24 +244,36 @@
}
public void start() {
+
+ // TODO: write a JUnit test for this problem (long-lasting or infinite loops because
+ // redundant children of a node are called recursively after when the node is extended
+ // twice)
/*
// String conceptStr = "(\"http://dl-learner.org/carcinogenesis#Compound\" AND (>= 2 \"http://dl-learner.org/carcinogenesis#hasStructure\".\"http://dl-learner.org/carcinogenesis#Ar_halide\" OR ((\"http://dl-learner.org/carcinogenesis#amesTestPositive\" IS TRUE) AND >= 5 \"http://dl-learner.org/carcinogenesis#hasBond\". TOP)))";
- String conceptStr = "(\"http://dl-learner.org/carcinogenesis#Compound\" AND ((\"http://dl-learner.org/carcinogenesis#amesTestPositive\" IS TRUE) AND (\"http://dl-learner.org/carcinogenesis#amesTestPositive\" IS TRUE)))";
+// String conceptStr = "(\"http://dl-learner.org/carcinogenesis#Compound\" AND ((\"http://dl-learner.org/carcinogenesis#amesTestPositive\" IS TRUE) AND (\"http://dl-learner.org/carcinogenesis#amesTestPositive\" IS TRUE)))";
+ String conceptStr = "(\"http://dl-learner.org/carcinogenesis#Compound\" AND (>= 3 \"http://dl-learner.org/carcinogenesis#hasStructure\".\"http://dl-learner.org/carcinogenesis#Halide\" OR ((\"http://dl-learner.org/carcinogenesis#amesTestPositive\" IS TRUE) AND ALL \"http://dl-learner.org/carcinogenesis#hasAtom\".TOP)))";
+ String conceptStr2 = "(\"http://dl-learner.org/carcinogenesis#Compound\" AND (>= 4 \"http://dl-learner.org/carcinogenesis#hasStructure\".\"http://dl-learner.org/carcinogenesis#Halide\" OR ((\"http://dl-learner.org/carcinogenesis#amesTestPositive\" IS TRUE) AND ALL \"http://dl-learner.org/carcinogenesis#hasAtom\".TOP)))";
try {
NamedClass struc = new NamedClass("http://dl-learner.org/carcinogenesis#Compound");
Description d = KBParser.parseConcept(conceptStr);
+ Description d2 = KBParser.parseConcept(conceptStr2);
// SortedSet<Description> ds = (SortedSet<Description>) operator.refine(d,15,null,struc);
// System.out.println(ds);
- System.out.println(RhoDRDown.checkIntersection((Intersection)d));
+// System.out.println(RhoDRDown.checkIntersection((Intersection)d));
Set<Individual> coveredNegatives = rs.instanceCheck(d, learningProblem.getNegativeExamples());
Set<Individual> coveredPositives = rs.instanceCheck(d, learningProblem.getPositiveExamples());
ExampleBasedNode ebn = new ExampleBasedNode(d);
ebn.setCoveredExamples(coveredPositives, coveredNegatives);
- extendNodeProper(ebn,15);
+ properRefinements.add(d2);
+ extendNodeProper(ebn,13);
+ extendNodeProper(ebn,14);
+ for(Description refinement: ebn.getChildConcepts())
+ System.out.println("refinement: " + refinement);
+
// Individual i = new Individual("http://dl-learner.org/carcinogenesis#d101");
// for(Individual i : learningProblem.getPositiveExamples())
// rs.instanceCheck(ds.last(), i);
@@ -308,7 +320,7 @@
// print statistics at most once a second
currentTime = System.nanoTime();
- if(currentTime - lastPrintTime > 1000000000) {
+ if(currentTime - lastPrintTime > 3000000000l) {
printStatistics(false);
lastPrintTime = currentTime;
logger.debug("--- loop " + loop + " started ---");
@@ -327,7 +339,7 @@
// Logger.getRootLogger().setLevel(Level.TRACE);
}
- System.out.println("next expanded: " + candidates.last().getShortDescription(nrOfPositiveExamples, nrOfNegativeExamples, baseURI));
+// System.out.println("next expanded: " + candidates.last().getShortDescription(nrOfPositiveExamples, nrOfNegativeExamples, baseURI));
// chose best node according to heuristics
bestNode = candidates.last();
// extend best node
@@ -434,8 +446,11 @@
childConceptsDeletionTimeNs += System.nanoTime() - childConceptsDeletionTimeNsStart;
-// if(refinements.size()<30)
-// System.out.println("refinements: " + refinements);
+// if(refinements.size()<30) {
+//// System.out.println("refinements: " + refinements);
+// for(Description refinement: refinements)
+// System.out.println("refinement: " + refinement);
+// }
long evaluateSetCreationTimeNsStart = System.nanoTime();
@@ -463,7 +478,7 @@
propernessTestsAvoidedByShortConceptConstruction++;
propernessDetected = true;
- System.out.println("refinement " + refinement + " can be shortened");
+// System.out.println("refinement " + refinement + " can be shortened");
// System.exit(0);
}
}
@@ -517,7 +532,7 @@
}
evaluateSetCreationTimeNs += System.nanoTime() - evaluateSetCreationTimeNsStart;
-// System.out.println("intermediate 1");
+// System.out.println("intermediate 1 " + node.getShortDescription(nrOfPositiveExamples, nrOfNegativeExamples, baseURI));
// System.out.println(toEvaluateConcepts.size());
@@ -678,8 +693,17 @@
for(Description refinement : refinements) {
// for(int i=0; i<=recDepth; i++)
// System.out.print(" ");
-// System.out.println("call: " + refinement + " [maxLength " + maxLength + "]");
- extendNodeProper(node, refinement, maxLength, recDepth+1);
+// System.out.println("call: " + refinement + " [maxLength " + maxLength + ", rec depth " + recDepth + "]");
+
+ // check for redundancy (otherwise we may run into very time-intensive loops,
+ // see planned JUnit test case $x)
+
+ long redundancyCheckTimeNsStart = System.nanoTime();
+ boolean redundant = properRefinements.contains(refinement);
+ redundancyCheckTimeNs += System.nanoTime() - redundancyCheckTimeNsStart;
+
+ if(!redundant)
+ extendNodeProper(node, refinement, maxLength, recDepth+1);
// for(int i=0; i<=recDepth; i++)
// System.out.print(" ");
// System.out.println("finished: " + refinement + " [maxLength " + maxLength + "]");
Modified: trunk/src/dl-learner/org/dllearner/algorithms/refexamples/MultiHeuristic.java
===================================================================
--- trunk/src/dl-learner/org/dllearner/algorithms/refexamples/MultiHeuristic.java 2008-03-17 10:21:51 UTC (rev 714)
+++ trunk/src/dl-learner/org/dllearner/algorithms/refexamples/MultiHeuristic.java 2008-03-19 08:11:19 UTC (rev 715)
@@ -79,6 +79,9 @@
private double gainBonusFactor;
private double nodeChildPenalty = 0.0001;
private double startNodeBonus = 1.0;
+ // penalise errors on positive examples harder than on negative examples
+ // (positive weight = 1)
+ private double negativeWeight = 0.8;
// examples
private int nrOfNegativeExamples;
@@ -113,11 +116,11 @@
}
public double getNodeScore(ExampleBasedNode node) {
- double accuracy = getAccuracy(node.getCoveredPositives().size(),node.getCoveredNegatives().size());
+ double accuracy = getWeightedAccuracy(node.getCoveredPositives().size(),node.getCoveredNegatives().size());
ExampleBasedNode parent = node.getParent();
double gain = 0;
if(parent != null) {
- double parentAccuracy = getAccuracy(parent.getCoveredPositives().size(),parent.getCoveredNegatives().size());
+ double parentAccuracy = getWeightedAccuracy(parent.getCoveredPositives().size(),parent.getCoveredNegatives().size());
gain = accuracy - parentAccuracy;
} else {
accuracy += startNodeBonus;
@@ -126,9 +129,8 @@
return accuracy + gainBonusFactor * gain - expansionPenaltyFactor * he - nodeChildPenalty * node.getChildren().size();
}
- private double getAccuracy(int coveredPositives, int coveredNegatives) {
- return (coveredPositives + nrOfNegativeExamples - coveredNegatives)/(double)nrOfExamples;
-
+ private double getWeightedAccuracy(int coveredPositives, int coveredNegatives) {
+ return (coveredPositives + negativeWeight * (nrOfNegativeExamples - coveredNegatives))/(double)nrOfExamples;
}
public static double getNodeScore(ExampleBasedNode node, int nrOfPositiveExamples, int nrOfNegativeExamples) {
Modified: trunk/src/dl-learner/org/dllearner/examples/Carcinogenesis.java
===================================================================
--- trunk/src/dl-learner/org/dllearner/examples/Carcinogenesis.java 2008-03-17 10:21:51 UTC (rev 714)
+++ trunk/src/dl-learner/org/dllearner/examples/Carcinogenesis.java 2008-03-19 08:11:19 UTC (rev 715)
@@ -123,6 +123,9 @@
private static boolean learnCarcinogenic = true;
private static boolean useNewGroups = true;
+ private static boolean createPTE1Conf = false;
+ private static boolean createPTE2Conf = false;
+
/**
* @param args
* No arguments supported.
@@ -133,14 +136,13 @@
public static void main(String[] args) throws FileNotFoundException, IOException,
ParseException {
- // TODO: newgroups are not mapped currently
String[] files = new String[] { "newgroups.pl", "ames.pl", "atoms.pl", "bonds.pl", "gentoxprops.pl",
"ind_nos.pl", "ind_pos.pl"};
// "pte2/canc_nos.pl", "pte2/pte2ames.pl", "pte2/pte2atoms.pl",
// "pte2/pte2bonds.pl", "pte2/pte2gentox.pl", "pte2/pte2ind_nos.pl", "pte2/pte2newgroups.pl"
// "train.b" => not a pure Prolog file but Progol/Aleph specific
// };
- File owlFile = new File("examples/carcinogenesis/pte.owl");
+ File owlFile = new File("examples/carcinogenesis/carcinogenesis.owl");
Program program = null;
long startTime, duration;
@@ -281,7 +283,6 @@
// generating test examples for PTE-1
// => put all in one file, because they were used as training for PTE-2
File confPTE1File = new File("examples/carcinogenesis/testpte1.conf");
- Files.clearFile(confPTE1File);
File testPTE1Positives = new File(prologDirectory + "pte1.f");
File testPTE1Negatives = new File(prologDirectory + "pte1.n");
@@ -289,16 +290,20 @@
List<Individual> negPTE1Examples = getExamples(testPTE1Negatives);
appendPosExamples(confTrainFile, posPTE1Examples);
appendNegExamples(confTrainFile, negPTE1Examples);
- Files.clearFile(confPTE1File);
- Files.appendFile(confPTE1File, "import(\"pte.owl\");\nreasoner=fastInstanceChecker;\n\n");
- appendPosExamples(confPTE1File, posPTE1Examples);
- appendNegExamples(confPTE1File, negPTE1Examples);
+ if(createPTE1Conf) {
+ Files.clearFile(confPTE1File);
+ Files.appendFile(confPTE1File, "import(\"pte.owl\");\nreasoner=fastInstanceChecker;\n\n");
+ appendPosExamples(confPTE1File, posPTE1Examples);
+ appendNegExamples(confPTE1File, negPTE1Examples);
+ }
// create a PTE-2 test file
- File confPTE2File = new File("examples/carcinogenesis/testpte2.conf");
- Files.clearFile(confPTE2File);
- Files.appendFile(confPTE2File, "import(\"pte.owl\");\nreasoner=fastInstanceChecker;\n\n");
- Files.appendFile(confPTE2File, getPTE2Examples());
+ if(createPTE2Conf) {
+ File confPTE2File = new File("examples/carcinogenesis/testpte2.conf");
+ Files.clearFile(confPTE2File);
+ Files.appendFile(confPTE2File, "import(\"pte.owl\");\nreasoner=fastInstanceChecker;\n\n");
+ Files.appendFile(confPTE2File, getPTE2Examples());
+ }
}
Modified: trunk/src/dl-learner/org/dllearner/reasoning/FastInstanceChecker.java
===================================================================
--- trunk/src/dl-learner/org/dllearner/reasoning/FastInstanceChecker.java 2008-03-17 10:21:51 UTC (rev 714)
+++ trunk/src/dl-learner/org/dllearner/reasoning/FastInstanceChecker.java 2008-03-19 08:11:19 UTC (rev 715)
@@ -325,10 +325,10 @@
return true;
// earyl abort: e.g. >= 10 hasStructure.Methyl;
// if there are 11 fillers and 2 are not Methyl, the result is false
- } /* else {
+ } else {
if(roleFillers.size() - index < number)
return false;
- }*/
+ }
}
return false;
} else if (description instanceof ObjectMaxCardinalityRestriction) {
@@ -363,10 +363,10 @@
return false;
// earyl abort: e.g. <= 5 hasStructure.Methyl;
// if there are 6 fillers and 2 are not Methyl, the result is true
- } /* else {
+ } else {
if(roleFillers.size() - index <= number)
return true;
- } */
+ }
}
return true;
} else if (description instanceof BooleanValueRestriction) {
Modified: trunk/src/dl-learner/org/dllearner/refinementoperators/RhoDRDown.java
===================================================================
--- trunk/src/dl-learner/org/dllearner/refinementoperators/RhoDRDown.java 2008-03-17 10:21:51 UTC (rev 714)
+++ trunk/src/dl-learner/org/dllearner/refinementoperators/RhoDRDown.java 2008-03-19 08:11:19 UTC (rev 715)
@@ -92,6 +92,10 @@
// maximum number of fillers for eeach role
private Map<ObjectProperty,Integer> maxNrOfFillers = new TreeMap<ObjectProperty,Integer>();
+ // limit for cardinality restrictions (this makes sense if we e.g. have compounds with up to
+ // more than 200 atoms but we are only interested in atoms with certain characteristics and do
+ // not want something like e.g. >= 204 hasAtom.NOT Carbon-87; which blows up the search space
+ private int cardinalityLimit = 5;
// start concept (can be used to start from an arbitrary concept, needs
// to be Thing or NamedClass), note that when you use e.g. Compound as
@@ -105,7 +109,7 @@
private int topRefinementsLength = 0;
private Map<NamedClass, Integer> topARefinementsLength = new TreeMap<NamedClass, Integer>();
// M is finite and this value is the maximum length of any value in M
- private static int mMaxLength = 3;
+ private static int mMaxLength = 4;
// the sets M_\top and M_A
private Map<Integer,SortedSet<Description>> m = new TreeMap<Integer,SortedSet<Description>>();
@@ -194,15 +198,22 @@
}
// determine the maximum number of fillers for each role
+ // (up to a specified cardinality maximum)
+ if(useCardinalityRestrictions) {
for(ObjectProperty op : rs.getAtomicRoles()) {
int maxFillers = 0;
Map<Individual,SortedSet<Individual>> opMembers = rs.getRoleMembers(op);
for(SortedSet<Individual> inds : opMembers.values()) {
if(inds.size()>maxFillers)
maxFillers = inds.size();
+ if(maxFillers >= cardinalityLimit) {
+ maxFillers = cardinalityLimit;
+ break;
+ }
}
maxNrOfFillers.put(op, maxFillers);
}
+ }
/*
String conceptStr = "(\"http://dl-learner.org/carcinogenesis#Compound\" AND (>= 2 \"http://dl-learner.org/carcinogenesis#hasStructure\".\"http://dl-learner.org/carcinogenesis#Ar_halide\" OR ((\"http://dl-learner.org/carcinogenesis#amesTestPositive\" IS TRUE) AND >= 5 \"http://dl-learner.org/carcinogenesis#hasBond\". TOP)))";
@@ -412,23 +423,41 @@
// rule 4: ALL r.D => <= (maxFillers-1) r.D
// (length increases by 1 so we have to check whether max length is sufficient)
- if(useCardinalityRestrictions) {
- if(maxLength > description.getLength() && maxNrOfFillers.get(ar)>1) {
- ObjectMaxCardinalityRestriction max = new ObjectMaxCardinalityRestriction(maxNrOfFillers.get(ar)-1,role,description.getChild(0));
- refinements.add(max);
- }
- }
+ // => commented out because this is acutally not a downward refinement
+// if(useCardinalityRestrictions) {
+// if(maxLength > description.getLength() && maxNrOfFillers.get(ar)>1) {
+// ObjectMaxCardinalityRestriction max = new ObjectMaxCardinalityRestriction(maxNrOfFillers.get(ar)-1,role,description.getChild(0));
+// refinements.add(max);
+// }
+// }
} else if (description instanceof ObjectCardinalityRestriction) {
+ ObjectPropertyExpression role = ((ObjectCardinalityRestriction)description).getRole();
+ Description range = opRanges.get(role);
+ int number = ((ObjectCardinalityRestriction)description).getCardinality();
if(description instanceof ObjectMaxCardinalityRestriction) {
- // <= x r.C => <= (x-1) r.C
+ // rule 1: <= x r.C => <= x r.D
+ tmp = refine(description.getChild(0), maxLength-3, null, range);
+
+ for(Description d : tmp) {
+ refinements.add(new ObjectMaxCardinalityRestriction(number,role,d));
+ }
+
+ // rule 2: <= x r.C => <= (x-1) r.C
ObjectMaxCardinalityRestriction max = (ObjectMaxCardinalityRestriction) description;
- int number = max.getNumber();
- if(number > 0)
+// int number = max.getNumber();
+ if(number > 1)
refinements.add(new ObjectMaxCardinalityRestriction(number-1,max.getRole(),max.getChild(0)));
+
} else if(description instanceof ObjectMinCardinalityRestriction) {
+ tmp = refine(description.getChild(0), maxLength-3, null, range);
+
+ for(Description d : tmp) {
+ refinements.add(new ObjectMinCardinalityRestriction(number,role,d));
+ }
+
// >= x r.C => >= (x+1) r.C
ObjectMinCardinalityRestriction min = (ObjectMinCardinalityRestriction) description;
- int number = min.getNumber();
+// int number = min.getNumber();
if(number < maxNrOfFillers.get(min.getRole()))
refinements.add(new ObjectMinCardinalityRestriction(number+1,min.getRole(),min.getChild(0)));
}
@@ -723,7 +752,7 @@
long mComputationTimeStartNs = System.nanoTime();
// initialise all possible lengths (1 to 3)
- for(int i=1; i<=3; i++) {
+ for(int i=1; i<=mMaxLength; i++) {
m.put(i, new TreeSet<Description>(conceptComparator));
}
@@ -777,6 +806,15 @@
m.put(3,m3);
+ SortedSet<Description> m4 = new TreeSet<Description>(conceptComparator);
+ if(useCardinalityRestrictions) {
+ for(ObjectProperty r : rs.getMostGeneralRoles()) {
+ int maxFillers = maxNrOfFillers.get(r);
+ m4.add(new ObjectMaxCardinalityRestriction(maxFillers-1, r, new Thing()));
+ }
+ }
+ m.put(4,m4);
+
mComputationTimeNs += System.nanoTime() - mComputationTimeStartNs;
}
@@ -866,6 +904,15 @@
mA.get(nc).put(3,m3);
+ SortedSet<Description> m4 = new TreeSet<Description>(conceptComparator);
+ if(useCardinalityRestrictions) {
+ for(ObjectProperty r : mgr.get(nc)) {
+ int maxFillers = maxNrOfFillers.get(r);
+ m4.add(new ObjectMaxCardinalityRestriction(maxFillers-1, r, new Thing()));
+ }
+ }
+ mA.get(nc).put(4,m4);
+
mComputationTimeNs += System.nanoTime() - mComputationTimeStartNs;
}
This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
|
