Skip to content

Commit

Permalink
#4150 - Upgrade dependencies
Browse files Browse the repository at this point in the history 
- opennlp 1.9.3 -> 2.3.0
- improve handling of adaptive features during prediction
  • Loading branch information
@reckart
reckart committed Sep 1, 2023
1 parent 7fb80ac commit 72bf21e
Show file tree
Hide file tree
Showing 6 changed files with 44 additions and 59 deletions.
2 changes: 1 addition & 1 deletion inception/inception-imls-opennlp/pom.xml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@
<artifactId>inception-imls-opennlp</artifactId>
<name>INCEpTION - ML - OpenNLP (v ${opennlp.version})</name>
<properties>
<opennlp.version>1.9.3</opennlp.version>
<opennlp.version>2.3.0</opennlp.version>
</properties>
<dependencies>
<dependency>
Expand Down
4 changes: 1 addition & 3 deletions ...udarmstadt/ukp/inception/recommendation/imls/opennlp/doccat/OpenNlpDoccatRecommender.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,6 @@
import java.util.ArrayList;
import java.util.Collection;
import java.util.List;
import java.util.Map;
import java.util.Map.Entry;
import java.util.Objects;

Expand Down Expand Up @@ -252,8 +251,7 @@ private List<DocumentSample> extractSamples(List<CAS> aCasses)
Type sampleUnitType = getType(cas, SAMPLE_UNIT);
Type tokenType = getType(cas, Token.class);

Map<AnnotationFS, List<AnnotationFS>> sampleUnits = indexCovered(cas, sampleUnitType,
tokenType);
var sampleUnits = indexCovered(cas, sampleUnitType, tokenType);
for (Entry<AnnotationFS, List<AnnotationFS>> e : sampleUnits.entrySet()) {
AnnotationFS sampleUnit = e.getKey();
Collection<AnnotationFS> tokens = e.getValue();
Expand Down
2 changes: 0 additions & 2 deletions ...tadt/ukp/inception/recommendation/imls/opennlp/doccat/OpenNlpDoccatRecommenderTraits.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,6 @@

import com.fasterxml.jackson.annotation.JsonIgnoreProperties;

import opennlp.tools.ml.AbstractTrainer;
import opennlp.tools.util.TrainingParameters;

@JsonIgnoreProperties(ignoreUnknown = true)
Expand Down Expand Up @@ -90,7 +89,6 @@ public void setPredictionLimit(int aPredictionLimit)
public TrainingParameters getParameters()
{
TrainingParameters parameters = TrainingParameters.defaultParams();
parameters.put(AbstractTrainer.VERBOSE_PARAM, false);
parameters.put(TrainingParameters.ITERATIONS_PARAM, iterations);
parameters.put(TrainingParameters.CUTOFF_PARAM, cutoff);
parameters.put(TrainingParameters.THREADS_PARAM, numThreads);
Expand Down
91 changes: 42 additions & 49 deletions ...a/de/tudarmstadt/ukp/inception/recommendation/imls/opennlp/ner/OpenNlpNerRecommender.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -32,8 +32,6 @@
import java.util.Objects;

import org.apache.uima.cas.CAS;
import org.apache.uima.cas.Feature;
import org.apache.uima.cas.Type;
import org.apache.uima.cas.text.AnnotationFS;
import org.apache.uima.jcas.tcas.Annotation;
import org.slf4j.Logger;
Expand All @@ -51,9 +49,7 @@
import de.tudarmstadt.ukp.inception.recommendation.api.recommender.RecommenderContext.Key;
import de.tudarmstadt.ukp.inception.recommendation.api.recommender.TrainingCapability;
import de.tudarmstadt.ukp.inception.rendering.model.Range;
import it.unimi.dsi.fastutil.ints.Int2ObjectMap;
import it.unimi.dsi.fastutil.ints.Int2ObjectOpenHashMap;
import it.unimi.dsi.fastutil.objects.Object2IntMap;
import it.unimi.dsi.fastutil.objects.Object2IntOpenHashMap;
import opennlp.tools.ml.BeamSearch;
import opennlp.tools.namefind.NameFinderME;
Expand Down Expand Up @@ -95,7 +91,7 @@ public boolean isReadyForPrediction(RecommenderContext aContext)
@Override
public void train(RecommenderContext aContext, List<CAS> aCasses) throws RecommendationException
{
List<NameSample> nameSamples = extractNameSamples(aCasses);
var nameSamples = extractNameSamples(aCasses);

if (nameSamples.size() < 2) {
aContext.warn("Not enough training data: [%d] items", nameSamples.size());
Expand Down Expand Up @@ -232,10 +228,11 @@ public EvaluationResult evaluate(List<CAS> aCasses, DataSplitter aDataSplitter)
// Evaluate
var labelPairs = new ArrayList<LabelPair>();
for (var sample : testSet) {
// clear adaptive data from feature generators if necessary
if (sample.isClearAdaptiveDataSet()) {
nameFinder.clearAdaptiveData();
}
// During evaluation, we sample data across documents and shuffle them into training and
// tests sets. Thus, we consider every sample as coming from a unique document and
// always clear the adaptive data between samples. clear adaptive data from feature
// generators if necessary
nameFinder.clearAdaptiveData();

// Span contains one NE, Array of them all in one sentence
var sentence = sample.getSentence();
Expand All @@ -259,14 +256,14 @@ private List<LabelPair> determineLabelsForASentence(String[] sentence, Span[] pr
int predictedNameIdx = 0;
int goldNameIdx = 0;

List<LabelPair> labelPairs = new ArrayList<>();
var labelPairs = new ArrayList<LabelPair>();
// Spans store which tokens are part of it as [begin,end).
// Tokens are counted 0 to length of sentence.
// Therefore go through all tokens, determine which span they are part of
// for predictions and gold ones. Assign label accordingly to the annotated-token.
for (int i = 0; i < sentence.length; i++) {

String predictedLabel = NO_NE_TAG;
var predictedLabel = NO_NE_TAG;
if (predictedNameIdx < predictedNames.length) {
Span predictedName = predictedNames[predictedNameIdx];
predictedLabel = determineLabel(predictedName, i);
Expand All @@ -276,7 +273,7 @@ private List<LabelPair> determineLabelsForASentence(String[] sentence, Span[] pr
}
}

String goldLabel = NO_NE_TAG;
var goldLabel = NO_NE_TAG;
if (goldNameIdx < goldNames.length) {
Span goldName = goldNames[goldNameIdx];
goldLabel = determineLabel(goldName, i);
Expand All @@ -297,7 +294,7 @@ private List<LabelPair> determineLabelsForASentence(String[] sentence, Span[] pr
*/
private String determineLabel(Span aName, int aTokenIdx)
{
String label = NO_NE_TAG;
var label = NO_NE_TAG;

if (aName.getStart() <= aTokenIdx && aName.getEnd() > aTokenIdx) {
label = aName.getType();
Expand All @@ -310,82 +307,78 @@ private List<NameSample> extractNameSamples(List<CAS> aCasses)
{
var nameSamples = new ArrayList<NameSample>();

casses: for (CAS cas : aCasses) {
Type sampleUnitType = getType(cas, SAMPLE_UNIT);
Type tokenType = getType(cas, Token.class);
nextCas: for (CAS cas : aCasses) {
var sampleUnitType = getType(cas, SAMPLE_UNIT);
var tokenType = getType(cas, Token.class);

for (AnnotationFS sampleUnit : cas.<Annotation> select(sampleUnitType)) {
var firstSampleInCas = true;
for (var sampleUnit : cas.<Annotation> select(sampleUnitType)) {
if (nameSamples.size() >= traits.getTrainingSetSizeLimit()) {
break casses;
break nextCas;
}

if (isBlank(sampleUnit.getCoveredText())) {
continue;
}

Collection<Annotation> tokens = cas.<Annotation> select(tokenType)
.coveredBy(sampleUnit).asList();

NameSample nameSample = createNameSample(cas, sampleUnit, tokens);
if (nameSample.getNames().length > 0) {
nameSamples.add(nameSample);
var tokens = cas.<Annotation> select(tokenType).coveredBy(sampleUnit).asList();
var tokenTexts = tokens.stream().map(AnnotationFS::getCoveredText)
.toArray(String[]::new);
var annotatedSpans = extractAnnotatedSpans(cas, sampleUnit, tokens);
if (annotatedSpans.length == 0) {
continue;
}

var nameSample = new NameSample(tokenTexts, annotatedSpans, firstSampleInCas);
nameSamples.add(nameSample);
firstSampleInCas = false;
}
}

return nameSamples;
}

private NameSample createNameSample(CAS aCas, AnnotationFS aSampleUnit,
Collection<? extends AnnotationFS> aTokens)
{
String[] tokenTexts = aTokens.stream().map(AnnotationFS::getCoveredText)
.toArray(String[]::new);
Span[] annotatedSpans = extractAnnotatedSpans(aCas, aSampleUnit, aTokens);
return new NameSample(tokenTexts, annotatedSpans, true);
}

private Span[] extractAnnotatedSpans(CAS aCas, AnnotationFS aSampleUnit,
Collection<? extends AnnotationFS> aTokens)
{
// Create spans from target annotations
Type annotationType = getType(aCas, layerName);
Feature feature = annotationType.getFeatureByBaseName(featureName);
List<AnnotationFS> annotations = selectCovered(annotationType, aSampleUnit);
var annotationType = getType(aCas, layerName);
var feature = annotationType.getFeatureByBaseName(featureName);
var annotations = selectCovered(annotationType, aSampleUnit);

if (annotations.isEmpty()) {
return new Span[0];
}

// Convert character offsets to token indices
Int2ObjectMap<AnnotationFS> idxTokenBeginOffset = new Int2ObjectOpenHashMap<>();
Int2ObjectMap<AnnotationFS> idxTokenEndOffset = new Int2ObjectOpenHashMap<>();
Object2IntMap<AnnotationFS> idxToken = new Object2IntOpenHashMap<>();
int idx = 0;
var idxTokenBeginOffset = new Int2ObjectOpenHashMap<AnnotationFS>();
var idxTokenEndOffset = new Int2ObjectOpenHashMap<AnnotationFS>();
var idxToken = new Object2IntOpenHashMap<AnnotationFS>();
var idx = 0;
for (AnnotationFS t : aTokens) {
idxTokenBeginOffset.put(t.getBegin(), t);
idxTokenEndOffset.put(t.getEnd(), t);
idxToken.put(t, idx);
idx++;
}

List<Span> result = new ArrayList<>();
int highestEndTokenPositionObserved = 0;
int numberOfAnnotations = annotations.size();
var result = new ArrayList<Span>();
var highestEndTokenPositionObserved = 0;
var numberOfAnnotations = annotations.size();
for (int i = 0; i < numberOfAnnotations; i++) {
AnnotationFS annotation = annotations.get(i);
String label = annotation.getFeatureValueAsString(feature);
var annotation = annotations.get(i);
var label = annotation.getFeatureValueAsString(feature);

AnnotationFS beginToken = idxTokenBeginOffset.get(annotation.getBegin());
AnnotationFS endToken = idxTokenEndOffset.get(annotation.getEnd());
var beginToken = idxTokenBeginOffset.get(annotation.getBegin());
var endToken = idxTokenEndOffset.get(annotation.getEnd());
if (beginToken == null || endToken == null) {
LOG.warn("Skipping annotation not starting/ending at token boundaries: [{}-{}, {}]",
annotation.getBegin(), annotation.getEnd(), label);
continue;
}

int begin = idxToken.getInt(beginToken);
int end = idxToken.getInt(endToken);
var begin = idxToken.getInt(beginToken);
var end = idxToken.getInt(endToken);

// If the begin offset of the current annotation is lower than the highest offset so far
// observed, then it is overlapping with some annotation that we have seen before.
Expand Down
2 changes: 0 additions & 2 deletions ...udarmstadt/ukp/inception/recommendation/imls/opennlp/ner/OpenNlpNerRecommenderTraits.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,6 @@
import com.fasterxml.jackson.annotation.JsonIgnore;
import com.fasterxml.jackson.annotation.JsonIgnoreProperties;

import opennlp.tools.ml.AbstractTrainer;
import opennlp.tools.util.TrainingParameters;

@JsonIgnoreProperties(ignoreUnknown = true)
Expand Down Expand Up @@ -70,7 +69,6 @@ public void setPredictionLimit(int aPredictionLimit)
public TrainingParameters getParameters()
{
TrainingParameters parameters = TrainingParameters.defaultParams();
parameters.put(AbstractTrainer.VERBOSE_PARAM, "false");
parameters.put(TrainingParameters.THREADS_PARAM, numThreads);
return parameters;
}
Expand Down
2 changes: 0 additions & 2 deletions ...udarmstadt/ukp/inception/recommendation/imls/opennlp/pos/OpenNlpPosRecommenderTraits.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,6 @@
import com.fasterxml.jackson.annotation.JsonIgnore;
import com.fasterxml.jackson.annotation.JsonIgnoreProperties;

import opennlp.tools.ml.AbstractTrainer;
import opennlp.tools.util.TrainingParameters;

@JsonIgnoreProperties(ignoreUnknown = true)
Expand Down Expand Up @@ -88,7 +87,6 @@ public void setTaggedTokensThreshold(double aTaggedTokensThreshold)
public TrainingParameters getParameters()
{
TrainingParameters parameters = TrainingParameters.defaultParams();
parameters.put(AbstractTrainer.VERBOSE_PARAM, "false");
parameters.put(TrainingParameters.THREADS_PARAM, numThreads);
return parameters;
}
Expand Down

0 comments on commit 72bf21e

Please sign in to comment.