Menu
▾
▴
[DL-Learner SVN] SF.net SVN: dl-learner:[3906] trunk
|
From: <lor...@us...> - 2013-02-27 13:49:04
|
Revision: 3906
http://dl-learner.svn.sourceforge.net/dl-learner/?rev=3906&view=rev
Author: lorenz_b
Date: 2013-02-27 13:48:57 +0000 (Wed, 27 Feb 2013)
Log Message:
-----------
Added cross validation to matching tests.
Modified Paths:
--------------
trunk/components-core/src/main/java/org/dllearner/kb/sparql/SparqlEndpoint.java
trunk/interfaces/src/main/java/org/dllearner/cli/CrossValidation.java
trunk/scripts/pom.xml
trunk/scripts/src/main/java/org/dllearner/scripts/OntologyMatching.java
trunk/scripts/src/test/java/org/dllearner/junit/OntologyMatchingTest.java
Modified: trunk/components-core/src/main/java/org/dllearner/kb/sparql/SparqlEndpoint.java
===================================================================
--- trunk/components-core/src/main/java/org/dllearner/kb/sparql/SparqlEndpoint.java 2013-02-25 12:24:29 UTC (rev 3905)
+++ trunk/components-core/src/main/java/org/dllearner/kb/sparql/SparqlEndpoint.java 2013-02-27 13:48:57 UTC (rev 3906)
@@ -210,6 +210,18 @@
return new SparqlEndpoint(u, defaultGraphURIs, new LinkedList<String>());
}
+ public static SparqlEndpoint getEndpointLOD2Cloud() {
+ URL u = null;
+ try {
+ u = new URL("http://lod.openlinksw.com/sparql/");
+ } catch (Exception e) {
+ e.printStackTrace();
+ }
+ LinkedList<String> defaultGraphURIs=new LinkedList<String>();
+// defaultGraphURIs.add("http://dbpedia.org");
+ return new SparqlEndpoint(u, defaultGraphURIs, new LinkedList<String>());
+ }
+
public static SparqlEndpoint getEndpointLinkedGeoData() {
URL u = null;
try {
Modified: trunk/interfaces/src/main/java/org/dllearner/cli/CrossValidation.java
===================================================================
--- trunk/interfaces/src/main/java/org/dllearner/cli/CrossValidation.java 2013-02-25 12:24:29 UTC (rev 3905)
+++ trunk/interfaces/src/main/java/org/dllearner/cli/CrossValidation.java 2013-02-27 13:48:57 UTC (rev 3906)
@@ -27,7 +27,9 @@
import java.util.List;
import java.util.Random;
import java.util.Set;
+import java.util.TreeSet;
+import org.dllearner.core.AbstractLearningProblem;
import org.dllearner.core.ComponentInitException;
import org.dllearner.core.AbstractCELA;
import org.dllearner.core.AbstractReasonerComponent;
@@ -35,6 +37,7 @@
import org.dllearner.core.owl.Individual;
import org.dllearner.learningproblems.Heuristics;
import org.dllearner.learningproblems.PosNegLP;
+import org.dllearner.learningproblems.PosOnlyLP;
import org.dllearner.utilities.Helper;
import org.dllearner.utilities.datastructures.Datastructures;
import org.dllearner.utilities.statistics.Stat;
@@ -70,7 +73,7 @@
}
- public CrossValidation(AbstractCELA la, PosNegLP lp, AbstractReasonerComponent rs, int folds, boolean leaveOneOut) {
+ public CrossValidation(AbstractCELA la, AbstractLearningProblem lp, AbstractReasonerComponent rs, int folds, boolean leaveOneOut) {
DecimalFormat df = new DecimalFormat();
@@ -81,11 +84,20 @@
List<Set<Individual>> testSetsNeg = new LinkedList<Set<Individual>>();
// get examples and shuffle them too
- Set<Individual> posExamples = ((PosNegLP)lp).getPositiveExamples();
+ Set<Individual> posExamples;
+ Set<Individual> negExamples;
+ if(lp instanceof PosNegLP){
+ posExamples = ((PosNegLP)lp).getPositiveExamples();
+ negExamples = ((PosNegLP)lp).getNegativeExamples();
+ } else if(lp instanceof PosOnlyLP){
+ posExamples = ((PosNegLP)lp).getPositiveExamples();
+ negExamples = new HashSet<Individual>();
+ } else {
+ throw new IllegalArgumentException("Only PosNeg and PosOnly learning problems are supported");
+ }
List<Individual> posExamplesList = new LinkedList<Individual>(posExamples);
+ List<Individual> negExamplesList = new LinkedList<Individual>(negExamples);
Collections.shuffle(posExamplesList, new Random(1));
- Set<Individual> negExamples = ((PosNegLP)lp).getNegativeExamples();
- List<Individual> negExamplesList = new LinkedList<Individual>(negExamples);
Collections.shuffle(negExamplesList, new Random(2));
// sanity check whether nr. of folds makes sense for this benchmark
@@ -137,8 +149,13 @@
Set<String> pos = Datastructures.individualSetToStringSet(trainingSetsPos.get(currFold));
Set<String> neg = Datastructures.individualSetToStringSet(trainingSetsNeg.get(currFold));
- lp.setPositiveExamples(trainingSetsPos.get(currFold));
- lp.setNegativeExamples(trainingSetsNeg.get(currFold));
+ if(lp instanceof PosNegLP){
+ ((PosNegLP)lp).setPositiveExamples(trainingSetsPos.get(currFold));
+ ((PosNegLP)lp).setNegativeExamples(trainingSetsNeg.get(currFold));
+ } else if(lp instanceof PosOnlyLP){
+ ((PosOnlyLP)lp).setPositiveExamples(new TreeSet<Individual>(trainingSetsPos.get(currFold)));
+ }
+
try {
lp.init();
Modified: trunk/scripts/pom.xml
===================================================================
--- trunk/scripts/pom.xml 2013-02-25 12:24:29 UTC (rev 3905)
+++ trunk/scripts/pom.xml 2013-02-27 13:48:57 UTC (rev 3906)
@@ -116,7 +116,7 @@
<dependency>
<groupId>net.sourceforge.owlapi</groupId>
<artifactId>owlapi-distribution</artifactId>
- <version>3.4</version>
+ <version>3.4.4-SNAPSHOT</version>
</dependency>
<dependency>
<groupId>net.sourceforge.owlapi</groupId>
Modified: trunk/scripts/src/main/java/org/dllearner/scripts/OntologyMatching.java
===================================================================
--- trunk/scripts/src/main/java/org/dllearner/scripts/OntologyMatching.java 2013-02-25 12:24:29 UTC (rev 3905)
+++ trunk/scripts/src/main/java/org/dllearner/scripts/OntologyMatching.java 2013-02-27 13:48:57 UTC (rev 3906)
@@ -44,6 +44,7 @@
import org.dllearner.learningproblems.PosOnlyLP;
import org.dllearner.reasoning.FastInstanceChecker;
import org.dllearner.reasoning.SPARQLReasoner;
+import org.dllearner.utilities.CrossValidation;
import org.dllearner.utilities.LabelShortFormProvider;
import org.dllearner.utilities.datastructures.Datastructures;
import org.dllearner.utilities.datastructures.SetManipulation;
@@ -75,6 +76,8 @@
import com.hp.hpl.jena.rdf.model.Statement;
import com.hp.hpl.jena.sparql.engine.http.QueryEngineHTTP;
import com.hp.hpl.jena.vocabulary.OWL;
+import com.hp.hpl.jena.vocabulary.RDF;
+import com.hp.hpl.jena.vocabulary.RDFS;
import com.jamonapi.Monitor;
import com.jamonapi.MonitorFactory;
@@ -95,6 +98,8 @@
private Map<Description, List<? extends EvaluatedDescription>> mappingKB2KB1;
private boolean posNegLearning = true;
+ private final boolean performCrossValidation = true;
+ private int fragmentDepth = 2;
/**
* The maximum number of positive examples, used for the SPARQL extraction and learning algorithm
@@ -118,6 +123,10 @@
this(new KnowledgeBase(endpoint1), new KnowledgeBase(endpoint2));
}
+ public void setFragmentDepth(int fragmentDepth) {
+ this.fragmentDepth = fragmentDepth;
+ }
+
public void start(){
mappingKB1KB2 = computeAlignment(kb1, kb2);
printMappingPretty(mappingKB1KB2);
@@ -283,14 +292,16 @@
fullFragment.add(positiveFragment);
fullFragment.add(negativeFragment);
- //here is the most difficult task, i.e. find a 'good' fragment of the KB on which we can learn
KnowledgeSource ks = convert(fullFragment);
//initialize the reasoner
+ logger.info("Initializing reasoner...");
AbstractReasonerComponent rc = new FastInstanceChecker(ks);
rc.init();
+ logger.info("Done.");
//initialize the learning problem
+ logger.info("Initializing learning problem...");
AbstractLearningProblem lp;
if(posNeg){
lp = new PosNegLPStandard(rc, positiveExamplesSample, negativeExamplesSample);
@@ -298,26 +309,34 @@
lp = new PosOnlyLP(rc, positiveExamplesSample);
}
lp.init();
+ logger.info("Done.");
- //apply the learning algorithm
- logger.info("Running learning algorithm...");
+ //initialize the learning algorithm
+ logger.info("Initializing learning algorithm...");
CELOE la = new CELOE(lp, rc);
la.setMaxExecutionTimeInSeconds(10);
la.setNoisePercentage(25);
la.init();
- la.start();
+ logger.info("Done.");
- try {
- QTL qtl = new QTL(lp, new LocalModelBasedSparqlEndpointKS(fullFragment));
- qtl.init();
- qtl.start();
- System.out.println(qtl.getSPARQLQuery());
- } catch (LearningProblemUnsupportedException e) {
- e.printStackTrace();
+ if(performCrossValidation){
+ CrossValidation cv = new CrossValidation(la, lp, rc, 5, false);
+ } else {
+ //apply the learning algorithm
+ logger.info("Running learning algorithm...");
+ la.start();
+ logger.info(la.getCurrentlyBestEvaluatedDescription());
}
-
- logger.info(la.getCurrentlyBestEvaluatedDescription());
+// try {
+// QTL qtl = new QTL(lp, new LocalModelBasedSparqlEndpointKS(fullFragment));
+// qtl.init();
+// qtl.start();
+// System.out.println(qtl.getSPARQLQuery());
+// } catch (LearningProblemUnsupportedException e) {
+// e.printStackTrace();
+// }
+
return la.getCurrentlyBestEvaluatedDescriptions(10);
} catch (ComponentInitException e) {
e.printStackTrace();
@@ -382,21 +401,32 @@
logger.info(i++ + "/" + size);
fullFragment.add(getFragment(ind, kb));
}
- //filter out triples with String literals, as there often occur are some syntax errors and they are not relevant for learning
+ filter(fullFragment);
+ return fullFragment;
+ }
+
+ private void filter(Model model) {
+ // filter out triples with String literals, as there often occur are
+ // some syntax errors and they are not relevant for learning
List<Statement> statementsToRemove = new ArrayList<Statement>();
- for(Iterator<Statement> iter = fullFragment.listStatements().toList().iterator(); iter.hasNext();){
+ for (Iterator<Statement> iter = model.listStatements().toList().iterator(); iter.hasNext();) {
Statement st = iter.next();
RDFNode object = st.getObject();
- if(object.isLiteral()){
-// statementsToRemove.add(st);
+ if (object.isLiteral()) {
+ // statementsToRemove.add(st);
Literal lit = object.asLiteral();
- if(lit.getDatatype() == null || lit.getDatatype().equals(XSD.STRING)){
+ if (lit.getDatatype() == null || lit.getDatatype().equals(XSD.STRING)) {
st.changeObject("shortened", "en");
- }
+ }
}
+ //remove statements like <x a owl:Class>
+ if(st.getPredicate().equals(RDF.type)){
+ if(object.equals(RDFS.Class.asNode()) || object.equals(OWL.Class.asNode()) || object.equals(RDFS.Literal.asNode())){
+ statementsToRemove.add(st);
+ }
+ }
}
- fullFragment.remove(statementsToRemove);
- return fullFragment;
+ model.remove(statementsToRemove);
}
/**
@@ -414,7 +444,7 @@
private Model getFragment(Individual ind, KnowledgeBase kb){
logger.debug("Loading fragment for " + ind.getName());
ConciseBoundedDescriptionGenerator cbdGen = new ConciseBoundedDescriptionGeneratorImpl(kb.getEndpoint(), kb.getCache());
- Model cbd = cbdGen.getConciseBoundedDescription(ind.getName(), 2);
+ Model cbd = cbdGen.getConciseBoundedDescription(ind.getName(), fragmentDepth);
logger.debug("Got " + cbd.size() + " triples.");
return cbd;
}
Modified: trunk/scripts/src/test/java/org/dllearner/junit/OntologyMatchingTest.java
===================================================================
--- trunk/scripts/src/test/java/org/dllearner/junit/OntologyMatchingTest.java 2013-02-25 12:24:29 UTC (rev 3905)
+++ trunk/scripts/src/test/java/org/dllearner/junit/OntologyMatchingTest.java 2013-02-27 13:48:57 UTC (rev 3906)
@@ -33,6 +33,8 @@
private KnowledgeBase worldFactBook;
private KnowledgeBase openCyc;
private KnowledgeBase linkedGeoData;
+
+ private final int fragmentDepth = 3;
@Before
public void setUp() throws Exception {
@@ -112,6 +114,7 @@
@Test
public void testSingleClassLinkedGeoDataToDBpedia() {
OntologyMatching matcher = new OntologyMatching(linkedGeoData, dbpedia);
+ matcher.setFragmentDepth(fragmentDepth);
NamedClass nc = new NamedClass("http://linkedgeodata.org/ontology/Aerodrome");
List<? extends EvaluatedDescription> mapping = matcher.computeMapping(nc, linkedGeoData, dbpedia);
Map<Description, List<? extends EvaluatedDescription>> alignment = new HashMap<Description, List<? extends EvaluatedDescription>>();
This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
|
