r65992 MediaWiki - Code Review archive

Repository:	MediaWiki
Revision:	< r65991‎ \| r65992 \| r65993 >
Date:	19:34, 6 May 2010
Author:	daniel
Status:	deferred
Tags:
Comment:	disambig testing and debugging (sliding coherence disambig - still not passing)
Modified paths:	/trunk/WikiWord/WikiWord/src/main/java/de/brightbyte/wikiword/disambig/CoherenceDisambiguator.java (modified) (history) /trunk/WikiWord/WikiWord/src/main/java/de/brightbyte/wikiword/disambig/Disambiguator.java (modified) (history) /trunk/WikiWord/WikiWord/src/main/java/de/brightbyte/wikiword/disambig/SlidingCoherenceDisambiguator.java (modified) (history) /trunk/WikiWord/WikiWord/src/test/java/de/brightbyte/wikiword/disambig/SlidingCoherenceDisambiguatorTest.java (modified) (history)

Diff [purge]

Index: trunk/WikiWord/WikiWord/src/test/java/de/brightbyte/wikiword/disambig/SlidingCoherenceDisambiguatorTest.java
—	—	@@ -7,6 +7,8 @@
8	8	import java.util.Map;
9	9
10	10	import de.brightbyte.data.Pair;
	11	+import de.brightbyte.io.ConsoleIO;
	12	+import de.brightbyte.io.Output;
11	13	import de.brightbyte.util.PersistenceException;
12	14	import de.brightbyte.wikiword.disambig.Disambiguator.Interpretation;
13	15	import de.brightbyte.wikiword.disambig.Disambiguator.Result;
—	—	@@ -16,6 +18,8 @@
17	19
18	20	public class SlidingCoherenceDisambiguatorTest extends DisambiguatorTestBase {
19	21
	22	+ private Output traceOutput = ConsoleIO.output;
	23	+
20	24	public SlidingCoherenceDisambiguatorTest() throws IOException, PersistenceException {
21	25	super();
22	26	}
—	—	@@ -42,7 +46,25 @@
43	47	assertTrue("UK as United_Kingdom", interpretations.contains( new Disambiguator.Interpretation<Term, LocalConcept>( uk_as_United_Kingdom )) );
44	48	assertTrue("UK as Great_Britain", interpretations.contains( new Disambiguator.Interpretation<Term, LocalConcept>( uk_as_Great_Britain )) );
45	49	assertTrue("UK as England", interpretations.contains( new Disambiguator.Interpretation<Term, LocalConcept>( uk_as_England )) );
	50	+
	51	+ ///////////////////////////////////////////////////////////////////////////////////
	52	+ Term freak = new Term("Freak");
	53	+ Pair<Term, LocalConcept> freak_as_nothing = new Pair<Term, LocalConcept>(freak, null);
46	54
	55	+ sequence = new ArrayList<Term>();
	56	+ sequence.add(freak);
	57	+ sequence.add(london);
	58	+
	59	+ interpretations = disambiguator.getSequenceInterpretations(sequence, meaningFetcher.getMeanings(sequence));
	60	+
	61	+ assertEquals("number of interpretations", 3, interpretations.size());
	62	+
	63	+ Interpretation<Term, LocalConcept> first = interpretations.iterator().next();
	64	+
	65	+ assertEquals( first.getSequence(), sequence );
	66	+ Interpretation<Term, LocalConcept> interp = new Disambiguator.Interpretation<Term, LocalConcept>( freak_as_nothing, london_as_City_of_London );
	67	+ assertTrue("London as City_of_London", interpretations.contains( interp) );
	68	+
47	69	///////////////////////////////////////////////////////////////////////////////////
48	70
49	71	sequence = new ArrayList<Term>();
—	—	@@ -103,6 +125,7 @@
104	126	PhraseOccuranceSet set = getBankAndMonumentPhrases();
105	127
106	128	SlidingCoherenceDisambiguator disambiguator = new SlidingCoherenceDisambiguator(meaningFetcher, featureFetcher);
	129	+ disambiguator.setTrace(traceOutput);
107	130	disambiguator.setInitialWindow(1);
108	131	disambiguator.setWindow(3);
109	132
—	—	@@ -115,13 +138,59 @@
116	139	assertEquals("Underground", sequence.get(1).getTerm());
117	140	assertEquals("station", sequence.get(2).getTerm());
118	141
119		~~- assertNotNull( meanings.get( sequence.get(0).getTerm() ) );~~
120		~~- assertNotNull( meanings.get( sequence.get(1).getTerm() ) );~~
121		~~- assertNotNull( meanings.get( sequence.get(2).getTerm() ) );~~
	142	+ assertNotNull( meanings.get( sequence.get(0) ) );
	143	+ assertNotNull( meanings.get( sequence.get(1) ) );
	144	+ assertNotNull( meanings.get( sequence.get(2) ) );
122	145
123		~~- assertEquals("Bank_and_Monument_Underground_station", meanings.get( sequence.get(0).getTerm() ).getName() );~~
124		~~- assertEquals("Subway", meanings.get( sequence.get(1).getTerm() ).getName() );~~
125		~~- assertEquals("Metro_station", meanings.get( sequence.get(2).getTerm() ).getName() );~~
	146	+ assertEquals("Bank_and_Monument_Underground_stations", meanings.get( sequence.get(0) ).getName() );
	147	+ assertEquals("London_Underground", meanings.get( sequence.get(1) ).getName() );
	148	+ assertEquals("Metro_station", meanings.get( sequence.get(2) ).getName() );
	149	+
	150	+ ///////////////////////////////////////////////////////////////////////////
	151	+ disambiguator.setTrace(traceOutput);
	152	+ disambiguator.setInitialWindow(2);
	153	+ disambiguator.setWindow(3);
	154	+
	155	+ result = disambiguator.disambiguate(set.getRootNode(), null);
	156	+
	157	+ sequence = result.getSequence();
	158	+ meanings = result.getMeanings();
	159	+
	160	+ assertEquals("Bank and Monument", sequence.get(0).getTerm());
	161	+ assertEquals("Underground", sequence.get(1).getTerm());
	162	+ assertEquals("station", sequence.get(2).getTerm());
	163	+
	164	+ assertNotNull( meanings.get( sequence.get(0) ) );
	165	+ assertNotNull( meanings.get( sequence.get(1) ) );
	166	+ assertNotNull( meanings.get( sequence.get(2) ) );
	167	+
	168	+ assertEquals("Bank_and_Monument_Underground_stations", meanings.get( sequence.get(0) ).getName() );
	169	+ assertEquals("London_Underground", meanings.get( sequence.get(1) ).getName() );
	170	+ assertEquals("Metro_station", meanings.get( sequence.get(2) ).getName() );
	171	+
	172	+ ///////////////////////////////////////////////////////////////////////////
	173	+ disambiguator.setTrace(traceOutput);
	174	+ disambiguator.setInitialWindow(3);
	175	+ disambiguator.setWindow(3);
	176	+
	177	+ result = disambiguator.disambiguate(set.getRootNode(), null);
	178	+
	179	+ sequence = result.getSequence();
	180	+ meanings = result.getMeanings();
	181	+
	182	+ assertEquals("Bank and Monument", sequence.get(0).getTerm());
	183	+ assertEquals("Underground", sequence.get(1).getTerm());
	184	+ assertEquals("station", sequence.get(2).getTerm());
	185	+
	186	+ assertNotNull( meanings.get( sequence.get(0) ) );
	187	+ assertNotNull( meanings.get( sequence.get(1) ) );
	188	+ assertNotNull( meanings.get( sequence.get(2) ) );
	189	+
	190	+ assertEquals("Bank_and_Monument_Underground_stations", meanings.get( sequence.get(0) ).getName() );
	191	+ assertEquals("London_Underground", meanings.get( sequence.get(1) ).getName() );
	192	+ assertEquals("Metro_station", meanings.get( sequence.get(2) ).getName() );
	193	+
	194	+ throw new UnsupportedOperationException("todo: window 1, 2, ...");
126	195	}
127	196
128	197	public void testDisambiguateTerms() throws PersistenceException {
Index: trunk/WikiWord/WikiWord/src/main/java/de/brightbyte/wikiword/disambig/CoherenceDisambiguator.java
—	—	@@ -284,24 +284,25 @@
285	285
286	286	Collection<Disambiguator.Interpretation<X, LocalConcept>> base = getSequenceInterpretations(sequence.subList(1, sequence.size()), meanings);
287	287
288		~~- if (m==null \|\| m.size()==0) return base;~~
289		-
290	288	List<Disambiguator.Interpretation<X, LocalConcept>> interpretations = new ArrayList<Disambiguator.Interpretation<X, LocalConcept>>();
291	289
292	290	for (Disambiguator.Interpretation<X, LocalConcept> be: base) {
293		~~- for (LocalConcept c: m) {~~
294		~~- Map<X, LocalConcept> e = new HashMap<X, LocalConcept>();~~
295		~~- e.putAll(be.getMeanings());~~
296		~~- e.put(t, c);~~
297		-
298		~~- if (!sequence.isEmpty()) {~~
	291	+ if (m==null \|\| m.isEmpty()) {
	292	+ Disambiguator.Interpretation<X, LocalConcept>interp = new Disambiguator.Interpretation<X, LocalConcept>(be.getMeanings(), sequence);
	293	+ interpretations.add(interp);
	294	+ } else {
	295	+ for (LocalConcept c: m) {
	296	+ Map<X, LocalConcept> e = new HashMap<X, LocalConcept>();
	297	+ e.putAll(be.getMeanings());
	298	+ e.put(t, c);
	299	+
299	300	Disambiguator.Interpretation<X, LocalConcept>interp = new Disambiguator.Interpretation<X, LocalConcept>(e, sequence);
300	301	interpretations.add(interp);
301	302	}
302	303	}
303	304	}
304	305
305		~~- trace(" ~ "+t+": "+m.size()+" meanings; collected "+interpretations.size()+" combinations");~~
	306	+ trace(" ~ "+t+": "+(m==null ? "no": m.size())+" meanings; collected "+interpretations.size()+" combinations");
306	307	return interpretations;
307	308	}
308	309
—	—	@@ -367,6 +368,10 @@
368	369	}
369	370
370	371	if (d<0) throw new IllegalArgumentException("encountered negative similarity score ("+d+") for "+a+" / "+b);
	372	+
	373	+ assert d>=0;
	374	+ assert d<=1;
	375	+
371	376	sim += d;
372	377	n ++; //should add up to interp.size*(combo.size()-1)/2, according to Gauss
373	378	}
—	—	@@ -384,14 +389,25 @@
385	390	}
386	391
387	392	//normalize
388		~~- sim = sim / n; //scale~~
389		~~- pop = pop / c; //scale~~
390		~~- weight = weight / c; //scale~~
	393	+ sim = n == 0 ? 0 : sim / n; //scale
	394	+ pop = c == 0 ? 0 : pop / c; //scale
	395	+ weight = c == 0 ? 0 : weight / c; //scale
391	396
	397	+ assert pop >= 0;
	398	+ assert sim >= 0;
	399	+ assert sim <= 1;
	400	+
392	401	double popf = popularityNormalizer.apply(pop);
393	402	double simf = similarityNormalizer.apply(sim);
	403	+
	404	+ assert popf>=0;
	405	+ assert popf<=1;
	406	+ assert simf>=0;
	407	+ assert simf<=1;
394	408
395	409	double score = scoreCombiner.apply(popf, simf);
	410	+ assert score>=0;
	411	+ assert score<=1;
396	412
397	413	return new Result<X, LocalConcept>(interp.getMeanings(), interp.getSequence(), score, "simf="+simf+", popf="+popf+", sim="+sim+", pop="+pop+", weight="+weight);
398	414	}
Index: trunk/WikiWord/WikiWord/src/main/java/de/brightbyte/wikiword/disambig/SlidingCoherenceDisambiguator.java
—	—	@@ -97,8 +97,11 @@
98	98
99	99	if (initialWindow > 0) { //apply full coherence disambig to initial window size. initialWindow == 1 will trigger a popularity disambig.
100	100	Collection<List<X>> sequences = getSequences(root, initialWindow);
101		~~- Result<X, LocalConcept> r = super.disambiguate(sequences, root, meanings, context);~~
	101	+ Result<X, LocalConcept> r;
102	102
	103	+ if (initialWindow == 1) r = popularityDisambiguator.disambiguate(sequences, root, meanings, context);
	104	+ else r = super.disambiguate(sequences, root, meanings, context);
	105	+
103	106	sequence.addAll(r.getSequence());
104	107	currentNode = getLastNode(root, sequence);
105	108	disambig.putAll(r.getMeanings());
—	—	@@ -112,7 +115,8 @@
113	116	PhraseNode<X> bestNode = null;
114	117
115	118	for (PhraseNode<X> n: successors) {
116		~~- Result<X, LocalConcept> r = evalStep(sequence, disambig, currentNode, meanings, context, similarities, features); //empty sequence will trigger popularity disambig~~
	119	+ Result<X, LocalConcept> r = evalStep(sequence, disambig, n, meanings, context, similarities, features); //empty sequence will trigger popularity disambig
	120	+ trace("evalStep("+n+"): " + r.toString());
117	121	if (best == null \|\| best.getScore() < r.getScore()) {
118	122	best = r;
119	123	bestNode = n;
Index: trunk/WikiWord/WikiWord/src/main/java/de/brightbyte/wikiword/disambig/Disambiguator.java
—	—	@@ -23,7 +23,7 @@
24	24	private static <T extends TermReference, C extends WikiWordConcept>Map<T, C> buildMeaningMap(List<Pair<T, C>> interpretation) {
25	25	Map<T, C> sequence = new HashMap<T, C>(interpretation.size());
26	26	for (Pair<T, C> p: interpretation) {
27		~~- sequence.put(p.getA(), p.getB());~~
	27	+ if (p.getB()!=null) sequence.put(p.getA(), p.getB());
28	28	}
29	29	return sequence;
30	30	}

Status & tagging log

15:15, 17 May 2010 MaxSem (talk | contribs) changed the status of r65992 [removed: new added: deferred]