Menu
▾
▴
[DL-Learner SVN] SF.net SVN: dl-learner:[3675] trunk/components-core/src/main/java/org/ dllearner
|
From: <lor...@us...> - 2012-05-02 15:25:24
|
Revision: 3675
http://dl-learner.svn.sourceforge.net/dl-learner/?rev=3675&view=rev
Author: lorenz_b
Date: 2012-05-02 14:52:45 +0000 (Wed, 02 May 2012)
Log Message:
-----------
Refactored measure for disjointness.
Added popularity precomputation for classes.
Modified Paths:
--------------
trunk/components-core/src/main/java/org/dllearner/algorithms/DisjointClassesLearner.java
trunk/components-core/src/main/java/org/dllearner/kb/sparql/SPARQLTasks.java
trunk/components-core/src/main/java/org/dllearner/reasoning/SPARQLReasoner.java
Modified: trunk/components-core/src/main/java/org/dllearner/algorithms/DisjointClassesLearner.java
===================================================================
--- trunk/components-core/src/main/java/org/dllearner/algorithms/DisjointClassesLearner.java 2012-05-02 09:26:50 UTC (rev 3674)
+++ trunk/components-core/src/main/java/org/dllearner/algorithms/DisjointClassesLearner.java 2012-05-02 14:52:45 UTC (rev 3675)
@@ -21,6 +21,7 @@
import java.net.URL;
import java.util.ArrayList;
+import java.util.Collections;
import java.util.HashMap;
import java.util.HashSet;
import java.util.List;
@@ -44,6 +45,7 @@
import org.dllearner.core.owl.NamedClass;
import org.dllearner.kb.LocalModelBasedSparqlEndpointKS;
import org.dllearner.kb.SparqlEndpointKS;
+import org.dllearner.kb.sparql.SparqlEndpoint;
import org.dllearner.learningproblems.AxiomScore;
import org.dllearner.learningproblems.Heuristics;
import org.slf4j.Logger;
@@ -84,6 +86,8 @@
private Set<NamedClass> allClasses;
+ private int popularity;
+
public DisjointClassesLearner(SparqlEndpointKS ks){
this.ks = ks;
}
@@ -119,9 +123,13 @@
fetchedRows = 0;
currentlyBestEvaluatedDescriptions = new ArrayList<EvaluatedDescription>();
- //TODO
+ //we return here if the class contains no instances
+ popularity = reasoner.getPopularity(classToDescribe);
+ if(popularity == 0){
+ return;
+ }
- //at first get all existing classes in knowledgebase
+ //at first get all existing classes in knowledge base
allClasses = getAllClasses();
allClasses.remove(classToDescribe);
@@ -316,17 +324,24 @@
//firstly, create disjoint classexpressions which not occur and give score of 1
for(NamedClass cls : completeDisjointclasses){
if(useClassPopularity){
- int popularity = 0;
+ int overlap = 0;
+ int pop;
if(ks.isRemote()){
- popularity = reasoner.getIndividualsCount(cls);
+ pop = reasoner.getPopularity(cls);
} else {
- popularity = ((LocalModelBasedSparqlEndpointKS)ks).getModel().getOntClass(cls.getName()).listInstances().toSet().size();
+ pop = ((LocalModelBasedSparqlEndpointKS)ks).getModel().getOntClass(cls.getName()).listInstances().toSet().size();
}
//we skip classes with no instances
- if(popularity == 0) continue;
- double[] confidenceInterval = Heuristics.getConfidenceInterval95Wald(popularity, 0);
- double accuracy = (confidenceInterval[0] + confidenceInterval[1]) / 2;
- evalDesc = new EvaluatedDescription(cls, new AxiomScore(1- accuracy));
+ if(pop == 0) continue;
+
+ //we compute the estimated precision
+ double precision = accuracy(pop, overlap);
+ //we compute the estimated recall
+ double recall = accuracy(popularity, overlap);
+ //compute the overall score
+ double score = 1 - fMEasure(precision, recall);
+
+ evalDesc = new EvaluatedDescription(cls, new AxiomScore(score));
} else {
evalDesc = new EvaluatedDescription(cls, new AxiomScore(1));
}
@@ -335,23 +350,51 @@
}
//secondly, create disjoint classexpressions with score 1 - (#occurence/#all)
- for(Entry<NamedClass, Integer> entry : sortByValues(class2Count)){
- //drop classes from OWL and RDF namespace
- if(entry.getKey().getName().startsWith(OWL2.getURI()) || entry.getKey().getName().startsWith(RDF.getURI()))continue;
-// evalDesc = new EvaluatedDescription(entry.getKey(),
-// new AxiomScore(1 - (entry.getValue() / (double)all)));
- double[] confidenceInterval = Heuristics.getConfidenceInterval95Wald(total, entry.getValue());
- double accuracy = (confidenceInterval[0] + confidenceInterval[1]) / 2;
- evalDesc = new EvaluatedDescription(entry.getKey(),
- new AxiomScore(1 - accuracy));
- evalDescs.add(evalDesc);
+ for (Entry<NamedClass, Integer> entry : sortByValues(class2Count)) {
+ NamedClass cls = entry.getKey();
+ // drop classes from OWL and RDF namespace
+ if (cls.getName().startsWith(OWL2.getURI()) || cls.getName().startsWith(RDF.getURI()))
+ continue;
+ if (useClassPopularity) {
+ int overlap = entry.getValue();
+ int pop;
+ if (ks.isRemote()) {
+ pop = reasoner.getPopularity(cls);
+ } else {
+ pop = ((LocalModelBasedSparqlEndpointKS) ks).getModel()
+ .getOntClass(cls.getName()).listInstances().toSet()
+ .size();
+ }
+ // we skip classes with no instances
+ if (pop == 0)
+ continue;
+
+ // we compute the estimated precision
+ double precision = accuracy(pop, overlap);
+ // we compute the estimated recall
+ double recall = accuracy(popularity, overlap);
+ // compute the overall score
+ double score = 1 - fMEasure(precision, recall);
+
+ evalDesc = new EvaluatedDescription(cls, new AxiomScore(score));
+ } else {
+ evalDesc = new EvaluatedDescription(cls, new AxiomScore(1));
+ }
}
class2Count.put(classToDescribe, total);
return evalDescs;
}
+ private double accuracy(int total, int success){
+ double[] confidenceInterval = Heuristics.getConfidenceInterval95Wald(total, success);
+ return (confidenceInterval[0] + confidenceInterval[1]) / 2;
+ }
+ private double fMEasure(double precision, double recall){
+ return 2 * precision * recall / (precision + recall);
+ }
+
private void keepMostGeneralClasses(Set<NamedClass> classes){
if(ks.isRemote()){
if(reasoner.isPrepared()){
@@ -384,12 +427,14 @@
}
public static void main(String[] args) throws Exception{
-// SparqlEndpointKS ks = new SparqlEndpointKS(new SparqlEndpoint(new URL("http://dbpedia.aksw.org:8902/sparql"), Collections.singletonList("http://dbpedia.org"), Collections.<String>emptyList()));
- SparqlEndpointKS ks = new LocalModelBasedSparqlEndpointKS(new URL("http://dl-learner.svn.sourceforge.net/viewvc/dl-learner/trunk/examples/swore/swore.rdf?revision=2217"));
+ SparqlEndpointKS ks = new SparqlEndpointKS(new SparqlEndpoint(new URL("http://dbpedia.aksw.org:8902/sparql"), Collections.singletonList("http://dbpedia.org"), Collections.<String>emptyList()));
+ ks = new LocalModelBasedSparqlEndpointKS(new URL("http://dl-learner.svn.sourceforge.net/viewvc/dl-learner/trunk/examples/swore/swore.rdf?revision=2217"));
+ ks = new SparqlEndpointKS(SparqlEndpoint.getEndpointDBpediaLiveAKSW());
DisjointClassesLearner l = new DisjointClassesLearner(ks);
- l.setClassToDescribe(new NamedClass("http://ns.softwiki.de/req/CustomerRequirement"));
+ l.setClassToDescribe(new NamedClass("http://dbpedia.org/ontology/Book"));
l.init();
-// l.getReasoner().prepareSubsumptionHierarchy();
+ l.getReasoner().prepareSubsumptionHierarchy();
+ l.getReasoner().precomputeClassPopularity();
// System.out.println(l.getReasoner().getClassHierarchy().getSubClasses(new NamedClass("http://dbpedia.org/ontology/Athlete"), false));System.exit(0);
l.start();
Modified: trunk/components-core/src/main/java/org/dllearner/kb/sparql/SPARQLTasks.java
===================================================================
--- trunk/components-core/src/main/java/org/dllearner/kb/sparql/SPARQLTasks.java 2012-05-02 09:26:50 UTC (rev 3674)
+++ trunk/components-core/src/main/java/org/dllearner/kb/sparql/SPARQLTasks.java 2012-05-02 14:52:45 UTC (rev 3675)
@@ -702,6 +702,10 @@
public Set<NamedClass> getAllClasses() {
Set<NamedClass> classes = new TreeSet<NamedClass>();
String query = "PREFIX owl: <http://www.w3.org/2002/07/owl#> SELECT ?c WHERE {?c a owl:Class} LIMIT 1000";
+ /*
+ * String query = "PREFIX owl: <http://www.w3.org/2002/07/owl#> PREFIX rdfs:<http://www.w3.org/2000/01/rdf-schema#> " +
+ "SELECT ?c WHERE {{?c a owl:Class} UNION {?c rdfs:subClassOf ?d} UNION {?d rdfs:subClassOf ?c}} LIMIT 1000";
+ */
SparqlQuery sq = new SparqlQuery(query, sparqlEndpoint);
ResultSet q = sq.send(false);
while (q.hasNext()) {
Modified: trunk/components-core/src/main/java/org/dllearner/reasoning/SPARQLReasoner.java
===================================================================
--- trunk/components-core/src/main/java/org/dllearner/reasoning/SPARQLReasoner.java 2012-05-02 09:26:50 UTC (rev 3674)
+++ trunk/components-core/src/main/java/org/dllearner/reasoning/SPARQLReasoner.java 2012-05-02 14:52:45 UTC (rev 3675)
@@ -93,7 +93,9 @@
private ClassHierarchy hierarchy;
private OntModel model;
+ private Map<NamedClass, Integer> classPopularityMap;
+
public SPARQLReasoner(SparqlEndpointKS ks) {
this.ks = ks;
@@ -111,6 +113,36 @@
this.model = model;
}
+ public void precomputeClassPopularity(){
+ logger.info("Precomputing class popularity ...");
+ classPopularityMap = new HashMap<NamedClass, Integer>();
+
+ Set<NamedClass> classes = new SPARQLTasks(ks.getEndpoint()).getAllClasses();
+ String queryTemplate = "SELECT (COUNT(*) AS ?cnt) WHERE {?s a <%s>}";
+
+ ResultSet rs;
+ for(NamedClass nc : classes){
+ rs = executeSelectQuery(String.format(queryTemplate, nc.getName()));
+ int cnt = rs.next().getLiteral("cnt").getInt();
+ classPopularityMap.put(nc, cnt);
+ }
+ }
+
+ public int getPopularity(NamedClass nc){
+ if(classPopularityMap.containsKey(nc)){
+ return classPopularityMap.get(nc);
+ } else {
+ System.out.println("Cache miss: " + nc);
+ String queryTemplate = "SELECT (COUNT(*) AS ?cnt) WHERE {?s a <%s>}";
+
+ ResultSet rs = executeSelectQuery(String.format(queryTemplate, nc.getName()));
+ int cnt = rs.next().getLiteral("cnt").getInt();
+ classPopularityMap.put(nc, cnt);
+ return cnt;
+ }
+
+ }
+
public final ClassHierarchy prepareSubsumptionHierarchy() {
logger.info("Preparing subsumption hierarchy ...");
long startTime = System.currentTimeMillis();
@@ -915,6 +947,10 @@
this.cache = cache;
}
+ public void setUseCache(boolean useCache) {
+ this.useCache = useCache;
+ }
+
private boolean executeAskQuery(String query){
boolean ret;
if(ks.isRemote()){
This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
|
