From a82815eb0521c12ba25c8018a19d0e44a3c1ab12 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?J=C3=A9r=C3=B4me=20Euzenat?= <Jerome.Euzenat@inria.fr>
Date: Thu, 15 Sep 2011 20:29:40 +0000
Subject: [PATCH] - Implemented extractss for ** alignments in
DistanceAlignment (impl) - Changed default type of StringDistAlignment to
"?*" for preserving the previous behaviour (impl) - Suppressed the guard for
applying the Hungarian algorithm to equisimilarity situation (impl) - Fixed a
bug on DistanceAlignment 1:1 extraction when the second ontology is larger
than the first one (impl) - Added corresponding test
---
html/relnotes.html | 14 +-
.../exmo/align/impl/DistanceAlignment.java | 246 +++++++++++-------
.../impl/method/StringDistAlignment.java | 13 +-
test/src/MatcherTest.java | 130 ++++++++-
4 files changed, 306 insertions(+), 97 deletions(-)
diff --git a/html/relnotes.html b/html/relnotes.html
index 57665a9c..8f9c27c0 100644
--- a/html/relnotes.html
+++ b/html/relnotes.html
@@ -56,12 +56,24 @@ The development of 4 versions continues.
<li>Add more tests</li>
</ul></p>
+<!-- NOTE FOR VERSION 5
+The default type of StringDistAlignment should be reverted to "**"
+with a warning:
+-->
+
<h2>Current SVN trunk version</h2>
-<!--h2>Version 4.3 (): xx/xx/xxxx - Zimt</h2-->
+<!--h2>Version 4.3 (): xx/xx/2011 - Zimt</h2-->
<p><ul compact="1">
<li>Full reengineering of the test generator (gen)</li>
+<li>Implemented <tt>extractss</tt> for ** alignments in <tt>DistanceAlignment</tt> (impl)</li>
+<li>Changed default type of <tt>StringDistAlignment</tt> to "?*" for
+ preserving the previous behaviour (impl)</li>
<li>Added level lines in the triangle display of <tt>GroupEval</tt> (util)</li>
+<li>Suppressed the guard for applying the Hungarian algorithm to
+ equisimilarity situations (impl)</li>
+<li>Fixed a bug on <tt>DistanceAlignment</tt> 1:1 extraction when the
+ second ontology is larger than the first one (impl)</li>
</ul></p>
<h2>Version 4.2 (1618): 31/05/2011 - Tring</h2>
diff --git a/src/fr/inrialpes/exmo/align/impl/DistanceAlignment.java b/src/fr/inrialpes/exmo/align/impl/DistanceAlignment.java
index 99864d18..76385ec7 100644
--- a/src/fr/inrialpes/exmo/align/impl/DistanceAlignment.java
+++ b/src/fr/inrialpes/exmo/align/impl/DistanceAlignment.java
@@ -1,7 +1,7 @@
/*
* $Id$
*
- * Copyright (C) INRIA, 2003-2010
+ * Copyright (C) INRIA, 2003-2011
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
@@ -138,16 +138,17 @@ public abstract class DistanceAlignment extends ObjectAlignment implements Align
if ( type.equals("?*") || type.equals("1*") || type.equals("?+") || type.equals("1+") ) return extractqs( threshold, params );
else if ( type.equals("??") || type.equals("1?") || type.equals("?1") || type.equals("11") ) return extractqq( threshold, params );
else if ( type.equals("*?") || type.equals("+?") || type.equals("*1") || type.equals("+1") ) return extractqs( threshold, params );
- else if ( type.equals("**") || type.equals("+*") || type.equals("*+") || type.equals("++") ) return extractqs( threshold, params );
+ else if ( type.equals("**") || type.equals("+*") || type.equals("*+") || type.equals("++") ) return extractss( threshold, params );
// The else should be an error message
else throw new AlignmentException("Unknown alignment type: "+type);
}
- // JE: It is now certainly possible to virtualise extraction has it has
+ // JE: It is now certainly possible to virtualise extraction as it has
// been done for printing matrix in MatrixMeasure (todo)
/**
* Extract the alignment of a ?* type
+ * Non symmetric: for each entity of onto1, take the highest if superior to threshold
* Complexity: O(n^2)
*/
@SuppressWarnings("unchecked") //ConcatenatedIterator
@@ -169,7 +170,7 @@ public abstract class DistanceAlignment extends ObjectAlignment implements Align
ontology2().getDataProperties().iterator());
for ( Object current : pit2 ){
if ( sim.getSimilarity() ) val = sim.getPropertySimilarity(prop1,current);
- else val = 1 - sim.getPropertySimilarity(prop1,current);
+ else val = 1. - sim.getPropertySimilarity(prop1,current);
if ( val > max) {
found = true; max = val; prop2 = current;
}
@@ -182,7 +183,7 @@ public abstract class DistanceAlignment extends ObjectAlignment implements Align
Object class2 = null;
for ( Object current : ontology2().getClasses() ) {
if ( sim.getSimilarity() ) val = sim.getClassSimilarity(class1,current);
- else val = 1 - sim.getClassSimilarity(class1,current);
+ else val = 1. - sim.getClassSimilarity(class1,current);
if (val > max) {
found = true; max = val; class2 = current;
}
@@ -213,10 +214,64 @@ public abstract class DistanceAlignment extends ObjectAlignment implements Align
return((Alignment)this);
}
+ /**
+ * Extract the alignment of a ** type
+ * Symmetric: return all elements above threshold
+ * Complexity: O(n^2)
+ */
+ @SuppressWarnings("unchecked") //ConcatenatedIterator
+ public Alignment extractss( double threshold, Properties params) {
+ double val = 0.;
+ try {
+ // Extract for properties
+ ConcatenatedIterator pit1 = new
+ ConcatenatedIterator(ontology1().getObjectProperties().iterator(),
+ ontology1().getDataProperties().iterator());
+ for( Object prop1 : pit1 ){
+ ConcatenatedIterator pit2 = new
+ ConcatenatedIterator(ontology2().getObjectProperties().iterator(),
+ ontology2().getDataProperties().iterator());
+ for ( Object prop2 : pit2 ){
+ if ( sim.getSimilarity() ) val = sim.getPropertySimilarity(prop1,prop2);
+ else val = 1. - sim.getPropertySimilarity(prop1,prop2);
+ if ( val > threshold ) addAlignCell(prop1,prop2, "=", val);
+ }
+ }
+ // Extract for classes
+ for ( Object class1 : ontology1().getClasses() ) {
+ for ( Object class2 : ontology2().getClasses() ) {
+ if ( sim.getSimilarity() ) val = sim.getClassSimilarity(class1,class2);
+ else val = 1. - sim.getClassSimilarity(class1,class2);
+ if (val > threshold ) addAlignCell(class1, class2, "=", val);
+ }
+ }
+ // Extract for individuals
+ if ( params.getProperty("noinst") == null ){
+ for ( Object ind1 : ontology1().getIndividuals() ) {
+ if ( ontology1().getEntityURI( ind1 ) != null ) {
+ for ( Object ind2 : ontology2().getIndividuals() ) {
+ if ( ontology2().getEntityURI( ind2 ) != null ) {
+ if ( sim.getSimilarity() ) val = sim.getIndividualSimilarity( ind1, ind2 );
+ else val = 1 - sim.getIndividualSimilarity( ind1, ind2 );
+ if ( val > threshold ) addAlignCell(ind1,ind2, "=", val);
+ }
+ }
+ }
+ }
+ }
+ } catch (OntowrapException owex) { owex.printStackTrace(); //}
+ } catch (AlignmentException alex) { alex.printStackTrace(); }
+ return((Alignment)this);
+ }
+
/**
* Extract the alignment of a ?? type
*
- * exact algorithm using the Hungarian method
+ * exact algorithm using the Hungarian method.
+ * This algorithm contains several guards to prevent the HungarianAlgorithm to
+ * raise problems:
+ * - It invert column and rows when nbrows > nbcol (Hungarian loops)
+ * - It prevents to generate alignments when one category has no elements.
*/
@SuppressWarnings("unchecked") //ConcatenatedIterator
public Alignment extractqq( double threshold, Properties params) {
@@ -227,29 +282,26 @@ public abstract class DistanceAlignment extends ObjectAlignment implements Align
// Create a matrix
int nbclasses1 = ontology1().nbClasses();
int nbclasses2 = ontology2().nbClasses();
- double[][] matrix = new double[nbclasses1][nbclasses2];
- Object[] class1 = new Object[nbclasses1];
- Object[] class2 = new Object[nbclasses2];
- int i = 0;
- for ( Object ob : ontology1().getClasses() ) {
- class1[i++] = ob;
- }
- int j = 0;
- for ( Object ob : ontology2().getClasses() ) {
- class2[j++] = ob;
- }
- // ival indicates if all cells have the same value, or different (-1)
- double ival = sim.getClassSimilarity(class1[0],class2[0]);
- for( i = 0; i < nbclasses1; i++ ){
- for( j = 0; j < nbclasses2; j++ ){
- if ( ival != -1. && ival != sim.getClassSimilarity(class1[i],class2[j] ) ) ival = -1.;
- if ( sim.getSimilarity() ) matrix[i][j] = sim.getClassSimilarity(class1[i],class2[j]);
- else matrix[i][j] = 1 - sim.getClassSimilarity(class1[i],class2[j]);
+ if ( nbclasses1 != 0 && nbclasses2 != 0 ) {
+ double[][] matrix = new double[nbclasses1][nbclasses2];
+ Object[] class1 = new Object[nbclasses1];
+ Object[] class2 = new Object[nbclasses2];
+ int i = 0;
+ for ( Object ob : ontology1().getClasses() ) {
+ class1[i++] = ob;
}
- }
- // Pass it to the algorithm
- if ( ival == -1. ) { // if values are the same, return an empty alignment
- int[][] result = HungarianAlgorithm.hgAlgorithm( matrix, "max" );
+ int j = 0;
+ for ( Object ob : ontology2().getClasses() ) {
+ class2[j++] = ob;
+ }
+ for( i = 0; i < nbclasses1; i++ ){
+ for( j = 0; j < nbclasses2; j++ ){
+ if ( sim.getSimilarity() ) matrix[i][j] = sim.getClassSimilarity(class1[i],class2[j]);
+ else matrix[i][j] = 1. - sim.getClassSimilarity(class1[i],class2[j]);
+ }
+ }
+ // Pass it to the algorithm
+ int[][] result = callHungarianMethod( matrix, nbclasses1, nbclasses2 );
// Extract the result
for( i=0; i < result.length ; i++ ){
// The matrix has been destroyed
@@ -264,56 +316,53 @@ public abstract class DistanceAlignment extends ObjectAlignment implements Align
}
}
}
- } catch (AlignmentException alex) { alex.printStackTrace(); }
- catch (OntowrapException owex) { owex.printStackTrace(); }
+ } catch ( AlignmentException alex) { alex.printStackTrace(); }
+ catch ( OntowrapException owex) { owex.printStackTrace(); }
// For properties
try{
int nbprop1 = ontology1().nbProperties();
int nbprop2 = ontology2().nbProperties();
- double[][] matrix = new double[nbprop1][nbprop2];
- Object[] prop1 = new Object[nbprop1];
- Object[] prop2 = new Object[nbprop2];
- int i = 0;
- ConcatenatedIterator pit1 = new
- ConcatenatedIterator(ontology1().getObjectProperties().iterator(),
- ontology1().getDataProperties().iterator());
- for ( Object ob: pit1 ) prop1[i++] = ob;
- int j = 0;
- ConcatenatedIterator pit2 = new
- ConcatenatedIterator(ontology2().getObjectProperties().iterator(),
- ontology2().getDataProperties().iterator());
- for ( Object ob: pit2 ) prop2[j++] = ob;
- double ival = sim.getPropertySimilarity(prop1[0],prop2[0]);
- for( i = 0; i < nbprop1; i++ ){
- for( j = 0; j < nbprop2; j++ ){
- if ( ival != -1. && ival != sim.getPropertySimilarity(prop1[i],prop2[j] ) ) ival = -1.;
- if ( sim.getSimilarity() ) matrix[i][j] = sim.getPropertySimilarity(prop1[i],prop2[j]);
- else
- matrix[i][j] = 1 - sim.getPropertySimilarity(prop1[i],prop2[j]);
+ if ( nbprop1 != 0 && nbprop2 != 0 ) {
+ double[][] matrix = new double[nbprop1][nbprop2];
+ Object[] prop1 = new Object[nbprop1];
+ Object[] prop2 = new Object[nbprop2];
+ int i = 0;
+ ConcatenatedIterator pit1 = new
+ ConcatenatedIterator(ontology1().getObjectProperties().iterator(),
+ ontology1().getDataProperties().iterator());
+ for ( Object ob: pit1 ) prop1[i++] = ob;
+ int j = 0;
+ ConcatenatedIterator pit2 = new
+ ConcatenatedIterator(ontology2().getObjectProperties().iterator(),
+ ontology2().getDataProperties().iterator());
+ for ( Object ob: pit2 ) prop2[j++] = ob;
+ for( i = 0; i < nbprop1; i++ ){
+ for( j = 0; j < nbprop2; j++ ){
+ if ( sim.getSimilarity() ) matrix[i][j] = sim.getPropertySimilarity(prop1[i],prop2[j]);
+ else matrix[i][j] = 1. - sim.getPropertySimilarity(prop1[i],prop2[j]);
+ }
}
- }
- // Pass it to the algorithm
- if ( ival == -1. ) {
- int[][] result = HungarianAlgorithm.hgAlgorithm( matrix, "max" );
- // Extract the result
- for( i=0; i < result.length ; i++ ){
- // The matrix has been destroyed
- double val;
- if ( sim.getSimilarity() ) val = sim.getPropertySimilarity(prop1[result[i][0]],prop2[result[i][1]]);
- else val = 1 - sim.getPropertySimilarity(prop1[result[i][0]],prop2[result[i][1]]);
- // JE: here using strict-> is a very good idea.
- // it means that alignments with 0. similarity
- // will be excluded from the best match.
- if( val > threshold ){
- addCell( new ObjectCell( (String)null, prop1[result[i][0]], prop2[result[i][1]], BasicRelation.createRelation("="), val ) );
+ // Pass it to the algorithm
+ int[][] result = callHungarianMethod( matrix, nbprop1, nbprop2 );
+ // Extract the result
+ for( i=0; i < result.length ; i++ ){
+ // The matrix has been destroyed
+ double val;
+ if ( sim.getSimilarity() ) val = sim.getPropertySimilarity(prop1[result[i][0]],prop2[result[i][1]]);
+ else val = 1 - sim.getPropertySimilarity(prop1[result[i][0]],prop2[result[i][1]]);
+ // JE: here using strict-> is a very good idea.
+ // it means that alignments with 0. similarity
+ // will be excluded from the best match.
+ if( val > threshold ){
+ addCell( new ObjectCell( (String)null, prop1[result[i][0]], prop2[result[i][1]], BasicRelation.createRelation("="), val ) );
+ }
}
}
- }
} catch (AlignmentException alex) { alex.printStackTrace(); }
catch (OntowrapException owex) { owex.printStackTrace(); }
// For individuals
if ( params.getProperty("noinst") == null ){
- try{
+ try {
// Create individual lists
Object[] ind1 = new Object[ontology1().nbIndividuals()];
Object[] ind2 = new Object[ontology2().nbIndividuals()];
@@ -331,40 +380,53 @@ public abstract class DistanceAlignment extends ObjectAlignment implements Align
ind1[nbind1++] = ob;
}
}
- double[][] matrix = new double[nbind1][nbind2];
- int i, j;
- double ival = sim.getIndividualSimilarity(ind1[0],ind2[0]);
- for( i=0; i < nbind1; i++ ){
- for( j=0; j < nbind2; j++ ){
- if ( ival != -1. && ival != sim.getIndividualSimilarity(ind1[i],ind2[j] ) ) ival = -1.;
- if ( sim.getSimilarity() ) matrix[i][j] = sim.getIndividualSimilarity(ind1[i],ind2[j]);
- else
- matrix[i][j] = 1 - sim.getIndividualSimilarity(ind1[i],ind2[j]);
+ if ( nbind1 != 0 && nbind2 != 0 ) {
+ double[][] matrix = new double[nbind1][nbind2];
+ int i, j;
+ for( i=0; i < nbind1; i++ ){
+ for( j=0; j < nbind2; j++ ){
+ if ( sim.getSimilarity() ) matrix[i][j] = sim.getIndividualSimilarity(ind1[i],ind2[j]);
+ else matrix[i][j] = 1 - sim.getIndividualSimilarity(ind1[i],ind2[j]);
+ }
}
- }
- // Pass it to the algorithm
- if ( ival == -1. ) {
- int[][] result = HungarianAlgorithm.hgAlgorithm( matrix, "max" );
- // Extract the result
- for( i=0; i < result.length ; i++ ){
- // The matrix has been destroyed
- double val;
- if ( sim.getSimilarity() ) val = sim.getIndividualSimilarity(ind1[result[i][0]],ind2[result[i][1]]);
- else val = 1 - sim.getIndividualSimilarity(ind1[result[i][0]],ind2[result[i][1]]);
- // JE: here using strict-> is a very good idea.
- // it means that alignments with 0. similarity
- // will be excluded from the best match.
- if( val > threshold ){
- addCell( new ObjectCell( (String)null, ind1[result[i][0]], ind2[result[i][1]], BasicRelation.createRelation("="), val ) );
+ // Pass it to the algorithm
+ int[][] result = callHungarianMethod( matrix, nbind1, nbind2 );
+ // Extract the result
+ for( i=0; i < result.length ; i++ ){
+ // The matrix has been destroyed
+ double val;
+ if ( sim.getSimilarity() ) val = sim.getIndividualSimilarity(ind1[result[i][0]],ind2[result[i][1]]);
+ else val = 1 - sim.getIndividualSimilarity(ind1[result[i][0]],ind2[result[i][1]]);
+ // JE: here using strict-> is a very good idea.
+ // it means that alignments with 0. similarity
+ // will be excluded from the best match.
+ if( val > threshold ){
+ addCell( new ObjectCell( (String)null, ind1[result[i][0]], ind2[result[i][1]], BasicRelation.createRelation("="), val ) );
+ }
}
}
- }
} catch (AlignmentException alex) { alex.printStackTrace(); //}
} catch (OntowrapException owex) { owex.printStackTrace(); }
}
return((Alignment)this);
}
+ public int[][] callHungarianMethod( double[][] matrix, int i, int j ) {
+ boolean transposed = false;
+ if ( i > j ) { // transposed aray (because rows>columns).
+ matrix = HungarianAlgorithm.transpose(matrix);
+ transposed = true;
+ }
+ int[][] result = HungarianAlgorithm.hgAlgorithm( matrix, "max" );
+ if ( transposed ) {
+ for( int k=0; k < result.length ; k++ ) {
+ int val = result[k][0]; result[k][0] = result[k][1]; result[k][1] = val;
+ }
+
+ }
+ return result;
+ }
+
/**
* Greedy algorithm:
* 1) dump the part of the matrix distance above threshold in a sorted set
diff --git a/src/fr/inrialpes/exmo/align/impl/method/StringDistAlignment.java b/src/fr/inrialpes/exmo/align/impl/method/StringDistAlignment.java
index e9ad98c8..39dc7813 100644
--- a/src/fr/inrialpes/exmo/align/impl/method/StringDistAlignment.java
+++ b/src/fr/inrialpes/exmo/align/impl/method/StringDistAlignment.java
@@ -1,7 +1,7 @@
/*
* $Id$
*
- * Copyright (C) INRIA, 2003-2010
+ * Copyright (C) INRIA, 2003-2011
*
* 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
@@ -80,10 +80,17 @@ public class StringDistAlignment extends DistanceAlignment implements AlignmentP
}
}
- /** Creation **/
+ /**
+ * Creation
+ * (4.3) For compatibility reason with previous versions, the type is set to
+ * "?*" so that the behaviour is the same.
+ * In future version (5.0), this should be reverted to "**",
+ * so the extractors will behave differently
+ **/
public StringDistAlignment() {
setSimilarity( new StringDistMatrixMeasure() );
- setType("**");
+ setType("?*");
+ //setType("11");
}
/* Processing */
diff --git a/test/src/MatcherTest.java b/test/src/MatcherTest.java
index 4eefd406..f38f2dac 100644
--- a/test/src/MatcherTest.java
+++ b/test/src/MatcherTest.java
@@ -1,7 +1,7 @@
/*
* $Id$
*
- * Copyright (C) INRIA, 2008-2010
+ * Copyright (C) INRIA, 2008-2011
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
@@ -34,9 +34,11 @@ import org.semanticweb.owl.align.AlignmentException;
import org.semanticweb.owl.align.AlignmentProcess;
import org.semanticweb.owl.align.Alignment;
import org.semanticweb.owl.align.Evaluator;
+import org.semanticweb.owl.align.Cell;
import fr.inrialpes.exmo.align.impl.renderer.RDFRendererVisitor;
import fr.inrialpes.exmo.align.impl.method.StringDistAlignment;
+import fr.inrialpes.exmo.align.impl.MatrixMeasure;
import fr.inrialpes.exmo.ontosim.string.StringDistances;
import fr.inrialpes.exmo.align.impl.eval.PRecEvaluator;
import fr.inrialpes.exmo.align.impl.URIAlignment;
@@ -117,4 +119,130 @@ $ java -jar lib/Procalign.jar file://$CWD/examples/rdf/edu.umbc.ebiquity.publica
alignment.init( new URI("file:examples/rdf/edu.umbc.ebiquity.publication.owl"), new URI("file:examples/rdf/edu.mit.visus.bibtex.owl"));
alignment.align( (Alignment)null, params );
}
+
+ /* This tests an errors in extraction methods (Hungarian method) */
+ @Test(groups = { "full", "impl", "raw" })
+ public void hungarianExtractionTest() throws Exception {
+ StringDistAlignment dal = new StringDistAlignment();
+ dal.init( new URI("file:examples/rdf/edu.umbc.ebiquity.publication.owl"), new URI("file:examples/rdf/edu.mit.visus.bibtex.owl") );
+ assertEquals( dal.nbCells(), 0 );
+ // Extract with nothing
+ Properties params = new Properties();
+ params.setProperty( "noinst", "1" );
+ dal.align( (Alignment)null, params ); // This initialises the matrix
+ assertEquals( dal.nbCells(), 10 );
+ // ****CLASSICAL EXTRACTIONS****
+ // Test ** extraction
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqs( 0., params );
+ assertEquals( dal.nbCells(), 10 );
+ // Test 11 extraction
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqq( 0., params );
+ assertEquals( dal.nbCells(), 10 );
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqqgreedy( 0., params );
+ assertEquals( dal.nbCells(), 10 );
+ // ****ZERO'ED ALIGNMENT****
+ // Do it with only one cell... (all 0. but 1)
+ MatrixMeasure mm = (MatrixMeasure)dal.getSimilarity();
+ for ( int i = mm.nbclass1-1; i >= 0; i-- ) {
+ for ( int j = mm.nbclass2-1; j >= 0; j-- ) {
+ mm.clmatrix[i][j] = 1.; // this is a distance...
+ }
+ }
+ // Useless
+ for ( int i = mm.nbprop1-1; i >= 0; i-- ) {
+ for ( int j = mm.nbprop2-1; j >= 0; j-- ) {
+ mm.prmatrix[i][j] = 1.; // distance...
+ }
+ }
+ //dal.printDistanceMatrix( params );
+ // Test ** extraction
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ //printAlignment( dal );
+ dal.extractqs( 0., params );
+ assertEquals( dal.nbCells(), 0 );
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqs( 1., params );
+ assertEquals( dal.nbCells(), 0 );
+ // Test 11 extraction
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqq( 0., params );
+ assertEquals( dal.nbCells(), 0 );
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqqgreedy( 0., params );
+ assertEquals( dal.nbCells(), 0 );
+ // ****ADDED ONE EXPECTED CORRESPONDENCE****
+ // Do it with only one cell... (all 0. but 1)
+ mm.clmatrix[5][5] = .5;
+ // Test ** extraction
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ //printAlignment( dal );
+ dal.extractqs( 0., params );
+ assertEquals( dal.nbCells(), 1 );
+ // ******** TEST THIS *******
+ //for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ //dal.extractqs( .6, params );
+ //assertEquals( dal.nbCells(), 0 );
+ // Test 11 extraction
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqq( 0., params );
+ assertEquals( dal.nbCells(), 1 );
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqqgreedy( 0., params );
+ assertEquals( dal.nbCells(), 1 );
+ // ****ADDED ONE EXPECTED CORRESPONDENCE****
+ // Do it with only one cell... (all 0. but 1)
+ mm.clmatrix[5][8] = .5;
+ mm.clmatrix[8][8] = .5;
+ mm.clmatrix[8][5] = .5;
+ // Test ** extraction
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ //printAlignment( dal );
+ dal.extractqs( 0., params );
+ assertEquals( dal.nbCells(), 2 );
+ // ******** TEST THIS *******
+ //for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ //dal.extractqs( .6, params );
+ //assertEquals( dal.nbCells(), 0 );
+ // Test 11 extraction
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqq( 0., params );
+ assertEquals( dal.nbCells(), 2 );
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqqgreedy( 0., params );
+ assertEquals( dal.nbCells(), 2 );
+ }
+
+ /* This identifies a very specific bug in the Hungarian algorithm when:
+ * class2 < class1
+ * a subset of class2 is matched with a subset of class1
+ * In a simple 0/1 alignment
+ */
+ @Test(groups = { "full", "impl", "raw" }, dependsOnMethods = { "hungarianExtractionTest" })
+ public void naughtyHungarianExtractionTest() throws Exception {
+ StringDistAlignment dal = new StringDistAlignment();
+ dal.init( new URI("file:examples/rdf/edu.mit.visus.bibtex.owl"), new URI("file:examples/rdf/edu.umbc.ebiquity.publication.owl"));
+ assertEquals( dal.nbCells(), 0 );
+ // Extract with nothing
+ Properties params = new Properties();
+ params.setProperty( "noinst", "1" );
+ dal.align( (Alignment)null, params ); // This initialises the matrix
+ assertEquals( dal.nbCells(), 10 );
+ // Test 11 extraction
+ for ( Cell c : dal ) { dal.removeAlignCell( c ); }
+ dal.extractqq( 0., params );
+ assertEquals( dal.nbCells(), 10 ); // **** java.lang.ArrayIndexOutOfBoundsException: 0
+ }
+
+
+ public void printAlignment( Alignment dal ) {
+ try {
+ for ( Cell c : dal ) {
+ System.err.println( "< "+c.getObject1AsURI(dal).getFragment()+" "+c.getRelation()+" "+c.getObject1AsURI(dal).getFragment()+" / "+c.getStrength() );
+ }
+ } catch (Exception e) {}
+ }
+
}
--
GitLab