You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@ctakes.apache.org by cl...@apache.org on 2017/06/19 18:38:39 UTC
svn commit: r1799256 - in
/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn:
ae/JointRelationTokenBasedAnnotator.java
eval/EvaluationOfNeuralJointRelations.java
Author: clin
Date: Mon Jun 19 18:38:39 2017
New Revision: 1799256
URL: http://svn.apache.org/viewvc?rev=1799256&view=rev
Log:
add a joint model for both EE and ET relations.
Added:
ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn/ae/JointRelationTokenBasedAnnotator.java
ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn/eval/EvaluationOfNeuralJointRelations.java
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn/ae/JointRelationTokenBasedAnnotator.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn/ae/JointRelationTokenBasedAnnotator.java?rev=1799256&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn/ae/JointRelationTokenBasedAnnotator.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn/ae/JointRelationTokenBasedAnnotator.java Mon Jun 19 18:38:39 2017
@@ -0,0 +1,536 @@
+package org.apache.ctakes.temporal.nn.ae;
+
+import java.io.BufferedReader;
+import java.io.BufferedWriter;
+import java.io.FileWriter;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.InputStreamReader;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.Collection;
+import java.util.HashMap;
+import java.util.HashSet;
+import java.util.Iterator;
+import java.util.List;
+import java.util.Map;
+import java.util.Map.Entry;
+import java.util.Set;
+
+import org.apache.ctakes.core.resource.FileLocator;
+import org.apache.ctakes.core.util.OntologyConceptUtil;
+import org.apache.ctakes.temporal.ae.TemporalRelationExtractorAnnotator.IdentifiedAnnotationPair;
+import org.apache.ctakes.temporal.nn.ae.EventTimeTokenBasedAnnotator.OutputMode;
+import org.apache.ctakes.temporal.nn.data.ArgContextProvider;
+import org.apache.ctakes.typesystem.type.relation.BinaryTextRelation;
+import org.apache.ctakes.typesystem.type.relation.RelationArgument;
+import org.apache.ctakes.typesystem.type.relation.TemporalTextRelation;
+import org.apache.ctakes.typesystem.type.syntax.BaseToken;
+import org.apache.ctakes.typesystem.type.textsem.EventMention;
+import org.apache.ctakes.typesystem.type.textsem.IdentifiedAnnotation;
+import org.apache.ctakes.typesystem.type.textsem.TimeMention;
+import org.apache.ctakes.typesystem.type.textspan.Sentence;
+import org.apache.uima.analysis_engine.AnalysisEngineProcessException;
+import org.apache.uima.fit.util.JCasUtil;
+import org.apache.uima.jcas.JCas;
+import org.apache.uima.jcas.tcas.Annotation;
+import org.cleartk.ml.CleartkAnnotator;
+import org.cleartk.ml.Feature;
+import org.cleartk.ml.Instance;
+import org.cleartk.util.ViewUriUtil;
+
+import com.google.common.collect.Lists;
+/**
+ * a joint annotator for annotating both event-time and event-event relations using neural models.
+ * @author chenlin
+ *
+ */
+public class JointRelationTokenBasedAnnotator extends CleartkAnnotator<String> {
+
+ public static final String NO_RELATION_CATEGORY = "none";
+ private static TimeMention coveringTimex;
+ public static Map<String, Integer> timex_idx;
+
+ //output modes:
+ //"TokenSeq": original token sequence
+ //"TokenTimeclass": original token sequence + timeClass tag
+ //"TokenTimeclassPosSeq": original token sequence + timeClass + pos seq tag
+ //"timeclass": <timex_timeClass>
+ //"timeclassPosSeq": <timex_timeClass_POStags>
+ //"posSeq": <timex_jj_nn>
+ //"singleTag": <timex>
+ //"indexTags": <timex_0>
+ public static OutputMode timexMode;
+
+ private BufferedReader reader;
+ private static FileWriter fstream;
+ private static BufferedWriter out;
+
+ public JointRelationTokenBasedAnnotator() {
+ timexMode = OutputMode.TokenSeq; //set the mode here
+ if(timexMode == OutputMode.IndexTags){
+ timex_idx = new HashMap<>();
+ }
+ }
+
+ public Map<String, Integer> TimexIdxReader(InputStream in) throws IOException{
+ reader = new BufferedReader(new InputStreamReader(in));
+ Map<String, Integer> timex_index = new HashMap<>();
+ String line = null;
+ while((line = reader.readLine()) != null){
+ line = line.trim();
+ int sep = line.lastIndexOf("|");
+ String timex = line.substring(0, sep);
+ Integer idx = Integer.parseInt(line.substring(sep+1));
+ timex_index.put(timex, idx);
+ }
+ reader.close();
+ return timex_index;
+ }
+
+ public static void TimexIdxWriter() throws IOException{
+ fstream = new FileWriter("target/eval/thyme/train_and_test/event-event/timex_idx.txt");
+ out = new BufferedWriter(fstream);
+
+ Iterator<Entry<String, Integer>> it = timex_idx.entrySet().iterator();
+ while (it.hasNext()) {
+
+ // the key/value pair is stored here in pairs
+ Map.Entry<String, Integer> pairs = it.next();
+ out.write(pairs.getKey() + "|" + pairs.getValue() + "\n");
+ }
+ out.close();
+ }
+ @Override
+ public void process(JCas jCas) throws AnalysisEngineProcessException {
+ if(timexMode == OutputMode.IndexTags && !this.isTraining()){
+ final String timexIdxMapFile = "target/eval/thyme/train_and_test/event-event/timex_idx.txt";
+ try {
+ timex_idx = TimexIdxReader(FileLocator.getAsStream(timexIdxMapFile));
+ } catch (IOException e) {
+ e.printStackTrace();
+ }
+ }
+
+ // Map<EventMention, Collection<EventMention>> coveringMap =
+ // JCasUtil.indexCovering(jCas, EventMention.class, EventMention.class);
+
+ // get all gold relation lookup
+ Map<List<Annotation>, BinaryTextRelation> relationLookup;
+ relationLookup = new HashMap<>();
+ if (this.isTraining()) {
+ relationLookup = new HashMap<>();
+ for (BinaryTextRelation relation : JCasUtil.select(jCas, BinaryTextRelation.class)) {
+ Annotation arg1 = relation.getArg1().getArgument();
+ Annotation arg2 = relation.getArg2().getArgument();
+ // The key is a list of args so we can do bi-directional lookup
+ List<Annotation> key = Arrays.asList(arg1, arg2);
+ if(relationLookup.containsKey(key)){
+ String reln = relationLookup.get(key).getCategory();
+ System.err.println("Error in: "+ ViewUriUtil.getURI(jCas).toString());
+ System.err.println("Error! This attempted relation " + relation.getCategory() +
+ " already has a relation " + reln + " at this span: " +
+ arg1.getCoveredText() + " -- " + arg2.getCoveredText());
+ } else {
+ relationLookup.put(key, relation);
+ }
+ }
+ }
+
+ for(Sentence sentence : JCasUtil.select(jCas, Sentence.class)) {
+ // collect all relevant relation arguments from the sentence
+ List<IdentifiedAnnotationPair> candidatePairs = getCandidateRelationArgumentPairs(jCas, sentence);
+
+ // walk through the pairs of annotations
+ for (IdentifiedAnnotationPair pair : candidatePairs) {
+ IdentifiedAnnotation arg1 = pair.getArg1();
+ IdentifiedAnnotation arg2 = pair.getArg2();
+
+ String context;
+ String cuis;
+ if(arg2.getBegin() < arg1.getBegin()) {
+ // ... event2 ... event1 ... scenario
+ if(timexMode == OutputMode.TokenSeq){
+ String arg2tag = "e2";
+ if(arg2 instanceof TimeMention){
+ arg2tag="t";
+ }
+ context = ArgContextProvider.getTokenContext(jCas, sentence, arg2, arg2tag, arg1, "e1", 2);
+
+ }else{
+ context = getTokenTimexContext(jCas, sentence, arg2, "e2", arg1, "e1", 2);
+ }
+ } else {
+ // ... event1 ... event2 ... scenario
+ if(timexMode == OutputMode.TokenSeq){
+ String arg2tag ="e2";
+ if(arg2 instanceof TimeMention){
+ arg2tag="t";
+ }
+ context = ArgContextProvider.getTokenContext(jCas, sentence, arg1, "e1", arg2, arg2tag, 2);
+ }else{
+ context = getTokenTimexContext(jCas, sentence, arg1, "e1", arg2, "e2", 2);
+ }
+ }
+
+ //get CUIs for two arguments
+ // Set<String> CUIs = getCuiDtrel(jCas, arg1);
+ // CUIs.addAll(getCuiDtrel(jCas, arg2));
+ //
+ // cuis = String.join(" ", CUIs);
+
+ //derive features based on context:
+ List<Feature> feats = new ArrayList<>();
+ String[] tokens = context.split(" ");
+ // String[] tokens = (context + "|" + cuis).split(" ");
+ for (String token: tokens){
+ feats.add(new Feature(token.toLowerCase()));
+ }
+
+ // during training, feed the features to the data writer
+ if(this.isTraining()) {
+ String category = getRelationCategory(relationLookup, arg1, arg2);
+
+ // drop some portion of negative examples during training
+ // if(category == null && coin.nextDouble() <= 0.5) {
+ // continue; // skip this negative example
+ // }
+
+ if(category == null) {
+ category = NO_RELATION_CATEGORY;
+ } else{
+ category = category.toLowerCase();
+ }
+ this.dataWriter.write(new Instance<>(category, feats));
+ } else {
+ String predictedCategory = this.classifier.classify(feats);
+
+ // add a relation annotation if a true relation was predicted
+ if (predictedCategory != null && !predictedCategory.equals(NO_RELATION_CATEGORY)) {
+
+ // if we predict an inverted relation, reverse the order of the
+ // arguments
+ //if for event-time relations:
+ if(arg1 instanceof TimeMention || arg2 instanceof TimeMention){
+ if(predictedCategory.endsWith("-1")) {
+ predictedCategory = predictedCategory.substring(0, predictedCategory.length() - 2);
+ if(arg1 instanceof TimeMention){
+ IdentifiedAnnotation temp = arg1;
+ arg1 = arg2;
+ arg2 = temp;
+ }
+ } else {
+ if(arg1 instanceof EventMention){
+ IdentifiedAnnotation temp = arg1;
+ arg1 = arg2;
+ arg2 = temp;
+ }
+ }
+
+ // createRelation(jCas, arg1, arg2, predictedCategory.toUpperCase(), 0.0);
+ }else{//if for event-event relations:
+ if (predictedCategory.endsWith("-1")) {
+ predictedCategory = predictedCategory.substring(0, predictedCategory.length() - 2);
+ IdentifiedAnnotation temp = arg1;
+ arg1 = arg2;
+ arg2 = temp;
+ }
+
+ // createRelation(jCas, arg1, arg2, predictedCategory.toUpperCase(), 0.0);
+ }
+
+ createRelation(jCas, arg1, arg2, predictedCategory.toUpperCase(), 0.0);
+ }
+ }
+ }
+
+ }
+ if(timexMode== OutputMode.IndexTags && !this.isTraining()){//in test time update the hashmap file for each cas
+ try {
+ TimexIdxWriter();
+ } catch (IOException e) {
+ // TODO Auto-generated catch block
+ e.printStackTrace();
+ }
+ }
+ }
+
+ private static Set<String> getCuiDtrel(JCas jCas, IdentifiedAnnotation arg) {
+ Set<String> CuiDtr = new HashSet<>();
+
+ List<EventMention> events = new ArrayList<>();
+ events.addAll(JCasUtil.selectCovering(jCas, EventMention.class, arg));
+ List<EventMention> realEvents = Lists.newArrayList();
+ String dtrel = null;
+ for( EventMention event : events){
+ if(!event.getClass().equals(EventMention.class)){//find all non-gold events
+ realEvents.add(event);
+ }else{
+ dtrel = event.getEvent().getProperties().getDocTimeRel();
+ }
+ }
+
+ if(dtrel != null && !dtrel.equals("AFTER")){
+ events = realEvents;
+ for(EventMention event : events){
+ for(String cui: OntologyConceptUtil.getCuis( event )){
+ CuiDtr.add(cui);
+ }
+ }
+ }
+ return CuiDtr;
+ }
+
+ // private static boolean hasBiggerSpan(EventMention event1, IdentifiedAnnotation arg1) {
+ // if(event1.getBegin()<=arg1.getBegin() && event1.getEnd()>arg1.getEnd()){
+ // return true;
+ // }
+ // if(event1.getBegin()<arg1.getBegin() && event1.getEnd()>=arg1.getEnd()){
+ // return true;
+ // }
+ // return false;
+ // }
+
+ public static String getTokenTimexContext(JCas jCas, Sentence sentence, IdentifiedAnnotation arg1, String leftType,
+ IdentifiedAnnotation arg2, String rightType, int contextSize) {
+ List<String> tokens = new ArrayList<>();
+ //select prior context:
+ List<TimeMention> preTimex = JCasUtil.selectCovered(jCas, TimeMention.class, sentence.getBegin(), arg1.getBegin());
+ List<TimeMention> betweenTimex = JCasUtil.selectCovered(jCas, TimeMention.class, arg1.getEnd(), arg2.getBegin());
+ List<TimeMention> afterTimex = JCasUtil.selectCovered(jCas, TimeMention.class, arg2.getEnd(), sentence.getEnd());
+
+ tokens = addTimex2TokenSequence(jCas, tokens, JCasUtil.selectPreceding(jCas, BaseToken.class, arg1, contextSize), preTimex, sentence);
+ //get arg1:
+ tokens.add("<" + leftType + ">");
+ if (arg1 instanceof TimeMention){
+ String timeTag = generateTimeTag(jCas, (TimeMention)arg1);
+ tokens.add(timeTag);
+ }else{
+ tokens.add(arg1.getCoveredText());
+ // tokens.add(getEventProperty((EventMention)arg1));
+ }
+ tokens.add("</" + leftType + ">");
+ tokens = addTimex2TokenSequence(jCas, tokens, JCasUtil.selectBetween(jCas, BaseToken.class, arg1, arg2), betweenTimex, sentence);
+ //arg2
+ tokens.add("<" + rightType + ">");
+ if (arg2 instanceof TimeMention){
+ String timeTag = generateTimeTag(jCas, (TimeMention)arg2);
+ tokens.add(timeTag);
+ }else{
+ tokens.add(arg2.getCoveredText());
+ // tokens.add(getEventProperty((EventMention)arg2));
+ }
+ tokens.add("</" + rightType + ">");
+ tokens = addTimex2TokenSequence(jCas, tokens, JCasUtil.selectFollowing(jCas, BaseToken.class, arg2, contextSize), afterTimex, sentence);
+
+
+ return String.join(" ", tokens).replaceAll("[\r\n]", " ");
+ }
+
+ private static String generateTimeTag(JCas jCas, TimeMention timex) {
+ String timeTag = "<timex";
+ if(timexMode == OutputMode.IndexTags){
+ timeTag = timeTag+"_";
+ int idx = 0;
+ String timeWord = timex.getCoveredText().toLowerCase().replaceAll("[\r\n]", " ");
+ if(timex_idx.containsKey(timeWord)){
+ idx = timex_idx.get(timeWord);
+ }else{
+ idx = timex_idx.size();
+ timex_idx.put(timeWord, idx);
+ }
+ timeTag = timeTag+ idx+">";
+ }else if(timexMode == OutputMode.Timeclass){
+ timeTag = "<timex_"+timex.getTimeClass()+">";
+ }else if(timexMode == OutputMode.TimeclassPosSeq){
+ timeTag = "<timex_"+timex.getTimeClass();
+ for(BaseToken token : JCasUtil.selectCovered(jCas, BaseToken.class, timex)){
+ timeTag = timeTag+"_"+token.getPartOfSpeech();
+ }
+ timeTag = timeTag+">";
+ }else if(timexMode == OutputMode.PosSeq){
+ timeTag = "<timex";
+ for(BaseToken token : JCasUtil.selectCovered(jCas, BaseToken.class, timex)){
+ timeTag = timeTag+"_"+token.getPartOfSpeech();
+ }
+ timeTag = timeTag+">";
+ }else if(timexMode == OutputMode.SingleTag){
+ timeTag = timeTag+">";
+ }else if(timexMode == OutputMode.TokenTimeclass){
+ timeTag = timex.getCoveredText()+" <timex_"+timex.getTimeClass()+">";
+ }else if(timexMode == OutputMode.TokenTimeclassPosSeq){
+ timeTag = timex.getCoveredText()+" <timex_"+timex.getTimeClass();
+ for(BaseToken token : JCasUtil.selectCovered(jCas, BaseToken.class, timex)){
+ timeTag = timeTag+"_"+token.getPartOfSpeech();
+ }
+ timeTag = timeTag+">";
+ }else if(timexMode == OutputMode.NoTag){
+ timeTag="";
+ }
+ return timeTag;
+ }
+
+ private static List<String> addTimex2TokenSequence(JCas jCas, List<String> tokens, List<BaseToken> tokenSequences, List<TimeMention> listOfTimex, Sentence sent) {
+ coveringTimex = null;
+ for (BaseToken baseToken : tokenSequences){
+ if(baseToken.getEnd() <= sent.getEnd() && sent.getBegin() <= baseToken.getBegin()){
+ TimeMention currentTimex = findCoveringTimex(baseToken, listOfTimex);
+ if(currentTimex == null){
+ if(coveringTimex != null ){
+ coveringTimex = null;
+ }
+ tokens.add(baseToken.getCoveredText());
+ }else{//current timex is not null
+ if( currentTimex != coveringTimex){
+ String timeTag = generateTimeTag(jCas, currentTimex);
+ tokens.add(timeTag);
+ coveringTimex = currentTimex;
+ }
+ }
+ }
+ }
+ return tokens;
+ }
+
+ private static TimeMention findCoveringTimex(BaseToken baseToken, List<TimeMention> timexs) {
+ for(TimeMention timex : timexs){
+ if(timex.getBegin()<= baseToken.getBegin() && timex.getEnd() >= baseToken.getEnd()){
+ return timex;
+ }
+ }
+ return null;
+ }
+
+
+ /**
+ * original way of getting label
+ * @param relationLookup
+ * @param arg1
+ * @param arg2
+ * @return
+ */
+ protected String getRelationCategory(
+ Map<List<Annotation>, BinaryTextRelation> relationLookup,
+ IdentifiedAnnotation arg1,
+ IdentifiedAnnotation arg2) {
+
+ BinaryTextRelation relation = relationLookup.get(Arrays.asList(arg1, arg2));
+ String category = null;
+ //for event-time relations:
+ if(arg1 instanceof TimeMention || arg2 instanceof TimeMention){
+
+ if (relation != null) {
+ category = relation.getCategory();
+ if(arg1 instanceof EventMention){
+ category = category + "-1";
+ }
+ } else {
+ relation = relationLookup.get(Arrays.asList(arg2, arg1));
+ if (relation != null) {
+ category = relation.getCategory();
+ if(arg2 instanceof EventMention){
+ category = category + "-1";
+ }
+ }
+ }
+ }else{
+
+ //for event-event relations:
+ if (relation != null) {
+ if (arg2.getBegin() < arg1.getBegin()) {
+ category = relation.getCategory() + "-1";
+ } else {
+ category = relation.getCategory();
+ }
+ } else {
+ relation = relationLookup.get(Arrays.asList(arg2, arg1));
+ if (relation != null) {
+ if(arg2.getBegin() < arg1.getBegin()){
+ category = relation.getCategory();
+ } else {
+ category = relation.getCategory() + "-1";
+ }
+ }
+ }
+ }
+
+ return category;
+ }
+
+ // protected String getRelationCategory2(Map<List<Annotation>, BinaryTextRelation> relationLookup,
+ // IdentifiedAnnotation arg1,
+ // IdentifiedAnnotation arg2) {
+ //
+ // // gold view representation (i.e. only contains relations)
+ // BinaryTextRelation arg1ContainsArg2 = relationLookup.get(Arrays.asList(arg1, arg2));
+ // BinaryTextRelation arg2ContainsArg1 = relationLookup.get(Arrays.asList(arg2, arg1));
+ //
+ // // now translate to position dependent representation (i.e. contains and contains-1)
+ // if(arg1ContainsArg2 != null) {
+ // // still need to know whether it's arg1 ... arg2 or arg2 ... arg1
+ // // because that determines whether it's contains or contains-1
+ // if(arg1.getBegin() < arg2.getBegin()) {
+ // return arg1ContainsArg2.getCategory();
+ // } else {
+ // return arg1ContainsArg2.getCategory() + "-1";
+ // }
+ // } else if(arg2ContainsArg1 != null) {
+ // if(arg1.getBegin() < arg2.getBegin()) {
+ // return arg2ContainsArg1.getCategory() + "-1";
+ // } else {
+ // return arg2ContainsArg1.getCategory();
+ // }
+ // } else {
+ // return null;
+ // }
+ // }
+
+ protected void createRelation(JCas jCas, IdentifiedAnnotation arg1,
+ IdentifiedAnnotation arg2, String predictedCategory, double confidence) {
+ RelationArgument relArg1 = new RelationArgument(jCas);
+ relArg1.setArgument(arg1);
+ relArg1.setRole("Arg1");
+ relArg1.addToIndexes();
+ RelationArgument relArg2 = new RelationArgument(jCas);
+ relArg2.setArgument(arg2);
+ relArg2.setRole("Arg2");
+ relArg2.addToIndexes();
+ TemporalTextRelation relation = new TemporalTextRelation(jCas);
+ relation.setArg1(relArg1);
+ relation.setArg2(relArg2);
+ relation.setCategory(predictedCategory);
+ relation.setConfidence(confidence);
+ relation.addToIndexes();
+ }
+
+ private static List<IdentifiedAnnotationPair> getCandidateRelationArgumentPairs(JCas jCas, Sentence sentence) {
+
+ List<IdentifiedAnnotationPair> pairs = Lists.newArrayList();
+ List<EventMention> events = new ArrayList<>(JCasUtil.selectCovered(jCas, EventMention.class, sentence));
+ // filter events
+ List<EventMention> realEvents = Lists.newArrayList();
+ for( EventMention event : events){
+ if(event.getClass().equals(EventMention.class)){
+ realEvents.add(event);
+ }
+ }
+ events = realEvents;
+
+ int eventNum = events.size();
+ for (int i = 0; i < eventNum-1; i++) {
+ for(int j = i+1; j < eventNum; j++) {
+ EventMention eventA = events.get(i);
+ EventMention eventB = events.get(j);
+ pairs.add(new IdentifiedAnnotationPair(eventA, eventB));
+ }
+ }
+
+ for (EventMention event: events){
+ for (TimeMention time : JCasUtil.selectCovered(jCas, TimeMention.class, sentence)) {
+ pairs.add(new IdentifiedAnnotationPair(event, time));
+ }
+ }
+
+ return pairs;
+ }
+}
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn/eval/EvaluationOfNeuralJointRelations.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn/eval/EvaluationOfNeuralJointRelations.java?rev=1799256&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn/eval/EvaluationOfNeuralJointRelations.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/nn/eval/EvaluationOfNeuralJointRelations.java Mon Jun 19 18:38:39 2017
@@ -0,0 +1,1065 @@
+/**
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+package org.apache.ctakes.temporal.nn.eval;
+
+import java.io.BufferedWriter;
+import java.io.File;
+import java.io.FileWriter;
+import java.io.PrintWriter;
+import java.net.URI;
+import java.util.ArrayDeque;
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.Collection;
+import java.util.Collections;
+import java.util.Deque;
+import java.util.List;
+import java.util.Map;
+import java.util.Set;
+
+import org.apache.ctakes.relationextractor.eval.RelationExtractorEvaluation.HashableArguments;
+import org.apache.ctakes.temporal.ae.DocTimeRelAnnotator;
+import org.apache.ctakes.temporal.ae.EventEventRelationAnnotator;
+import org.apache.ctakes.temporal.eval.EvaluationOfEventTimeRelations.ParameterSettings;
+import org.apache.ctakes.temporal.keras.KerasStringOutcomeDataWriter;
+import org.apache.ctakes.temporal.keras.ScriptStringFeatureDataWriter;
+import org.apache.ctakes.temporal.nn.ae.JointRelationTokenBasedAnnotator;
+import org.apache.ctakes.temporal.eval.EvaluationOfTemporalRelations_ImplBase;
+import org.apache.ctakes.temporal.eval.Evaluation_ImplBase;
+import org.apache.ctakes.temporal.eval.I2B2Data;
+import org.apache.ctakes.temporal.eval.THYMEData;
+//import org.apache.ctakes.temporal.eval.Evaluation_ImplBase.WriteI2B2XML;
+//import org.apache.ctakes.temporal.eval.Evaluation_ImplBase.XMLFormat;
+import org.apache.ctakes.temporal.utils.AnnotationIdCollection;
+import org.apache.ctakes.temporal.utils.TLinkTypeArray2;
+import org.apache.ctakes.typesystem.type.relation.BinaryTextRelation;
+import org.apache.ctakes.typesystem.type.relation.RelationArgument;
+import org.apache.ctakes.typesystem.type.relation.TemporalTextRelation;
+import org.apache.ctakes.typesystem.type.textsem.EventMention;
+import org.apache.ctakes.typesystem.type.textsem.IdentifiedAnnotation;
+import org.apache.ctakes.typesystem.type.textsem.TimeMention;
+import org.apache.ctakes.typesystem.type.textspan.Sentence;
+import org.apache.uima.analysis_engine.AnalysisEngineDescription;
+import org.apache.uima.analysis_engine.AnalysisEngineProcessException;
+import org.apache.uima.cas.CAS;
+import org.apache.uima.cas.CASException;
+import org.apache.uima.collection.CollectionReader;
+import org.apache.uima.fit.component.JCasAnnotator_ImplBase;
+import org.apache.uima.fit.descriptor.ConfigurationParameter;
+import org.apache.uima.fit.factory.AggregateBuilder;
+import org.apache.uima.fit.factory.AnalysisEngineFactory;
+import org.apache.uima.fit.pipeline.JCasIterator;
+import org.apache.uima.fit.pipeline.SimplePipeline;
+import org.apache.uima.fit.util.JCasUtil;
+import org.apache.uima.jcas.JCas;
+import org.apache.uima.jcas.tcas.Annotation;
+import org.apache.uima.resource.ResourceInitializationException;
+import org.apache.uima.util.FileUtils;
+import org.cleartk.eval.AnnotationStatistics;
+import org.cleartk.ml.CleartkAnnotator;
+import org.cleartk.ml.jar.DefaultDataWriterFactory;
+import org.cleartk.ml.jar.DirectoryDataWriterFactory;
+import org.cleartk.ml.jar.GenericJarClassifierFactory;
+import org.cleartk.ml.jar.JarClassifierBuilder;
+import org.cleartk.ml.liblinear.LibLinearStringOutcomeDataWriter;
+//import org.cleartk.ml.libsvm.tk.TkLibSvmStringOutcomeDataWriter;
+//import org.cleartk.ml.libsvm.LIBSVMStringOutcomeDataWriter;
+//import org.cleartk.ml.tksvmlight.TKSVMlightStringOutcomeDataWriter;
+import org.cleartk.ml.tksvmlight.model.CompositeKernel.ComboOperator;
+import org.cleartk.util.ViewUriUtil;
+
+import com.google.common.base.Function;
+import com.google.common.collect.HashMultimap;
+import com.google.common.collect.Lists;
+import com.google.common.collect.Maps;
+import com.google.common.collect.Multimap;
+import com.google.common.collect.Sets;
+import com.lexicalscope.jewel.cli.CliFactory;
+import com.lexicalscope.jewel.cli.Option;
+
+/**
+ * This is the evaluation code to evaluate a joint neural model for predicting temporal relations, including event-time and event-event relations.
+ * @author chenlin
+ *
+ */
+public class EvaluationOfNeuralJointRelations extends
+EvaluationOfTemporalRelations_ImplBase{
+ static interface TempRelOptions extends Evaluation_ImplBase.Options{
+ @Option
+ public boolean getPrintFormattedRelations();
+
+ @Option
+ public boolean getBaseline();
+
+ @Option
+ public boolean getClosure();
+
+ @Option
+ public boolean getUseTmp();
+
+ @Option
+ public boolean getUseGoldAttributes();
+
+ @Option
+ public boolean getSkipTrain();
+
+ @Option
+ public boolean getWriteProbabilities();
+
+ @Option
+ public boolean getTestOnTrain();
+
+ @Option
+ public boolean getSkipWrite();
+ }
+
+ // protected static boolean DEFAULT_BOTH_DIRECTIONS = false;
+ // protected static float DEFAULT_DOWNSAMPLE = 1.0f;
+ // private static double DEFAULT_SVM_C = 1.0;
+ // private static double DEFAULT_SVM_G = 1.0;
+ // private static double DEFAULT_TK = 0.5;
+ // private static double DEFAULT_LAMBDA = 0.5;
+
+ // defaultParams = new ParameterSettings(DEFAULT_BOTH_DIRECTIONS, DEFAULT_DOWNSAMPLE, "tk",
+ // DEFAULT_SVM_C, DEFAULT_SVM_G, "polynomial", ComboOperator.SUM, DEFAULT_TK, DEFAULT_LAMBDA);
+ protected static ParameterSettings flatParams = new ParameterSettings(DEFAULT_BOTH_DIRECTIONS, DEFAULT_DOWNSAMPLE, "linear",
+ 10.0, 1.0, "linear", ComboOperator.VECTOR_ONLY, DEFAULT_TK, DEFAULT_LAMBDA);
+ protected static ParameterSettings allBagsParams = new ParameterSettings(DEFAULT_BOTH_DIRECTIONS, DEFAULT_DOWNSAMPLE, "tk",
+ 100.0, 0.1, "radial basis function", ComboOperator.SUM, 0.5, 0.5);
+ protected static ParameterSettings allParams = new ParameterSettings(DEFAULT_BOTH_DIRECTIONS, DEFAULT_DOWNSAMPLE, "tk",
+ 10.0, 1.0, "polynomial", ComboOperator.SUM, 0.1, 0.5); // (0.3, 0.4 for tklibsvm)
+ protected static ParameterSettings ftParams = new ParameterSettings(DEFAULT_BOTH_DIRECTIONS, DEFAULT_DOWNSAMPLE, "tk",
+ 1.0, 0.1, "radial basis function", ComboOperator.SUM, 0.5, 0.5);
+ private static Boolean recallModeEvaluation = true;
+
+ public static void main(String[] args) throws Exception {
+ TempRelOptions options = CliFactory.parseArguments(TempRelOptions.class, args);
+ List<Integer> trainItems = null;
+ List<Integer> devItems = null;
+ List<Integer> testItems = null;
+
+ List<Integer> patientSets = options.getPatients().getList();
+ if(options.getXMLFormat() == XMLFormat.I2B2){
+ trainItems = I2B2Data.getTrainPatientSets(options.getXMLDirectory());
+ devItems = I2B2Data.getDevPatientSets(options.getXMLDirectory());
+ testItems = I2B2Data.getTestPatientSets(options.getXMLDirectory());
+ }else{
+ trainItems = THYMEData.getPatientSets(patientSets, options.getTrainRemainders().getList());
+ devItems = THYMEData.getPatientSets(patientSets, options.getDevRemainders().getList());
+ testItems = THYMEData.getPatientSets(patientSets, options.getTestRemainders().getList());
+ }
+ ParameterSettings params = allParams;
+
+ // possibleParams.add(defaultParams);
+
+ // for(ParameterSettings params : possibleParams){
+ try{
+ File workingDir = new File("/Volumes/chip-nlp/Public/THYME/eval/thyme/");//"target/eval/thyme/");
+ if(!workingDir.exists()) workingDir.mkdirs();
+ if(options.getUseTmp()){
+ File tempModelDir = File.createTempFile("temporal", null, workingDir);
+ tempModelDir.delete();
+ tempModelDir.mkdir();
+ workingDir = tempModelDir;
+ }
+ EvaluationOfNeuralJointRelations evaluation = new EvaluationOfNeuralJointRelations(
+ workingDir,
+ options.getRawTextDirectory(),
+ options.getXMLDirectory(),
+ options.getXMLFormat(),
+ options.getSubcorpus(),
+ options.getXMIDirectory(),
+ options.getTreebankDirectory(),
+ options.getClosure(),
+ options.getPrintErrors(),
+ options.getPrintFormattedRelations(),
+ options.getBaseline(),
+ options.getUseGoldAttributes(),
+ options.getKernelParams(),
+ params);
+ // evaluation.prepareXMIsFor(patientSets);
+ if(options.getI2B2Output()!=null) evaluation.setI2B2Output(options.getI2B2Output() + "/temporal-relations/joint");
+ if(options.getAnaforaOutput()!=null) evaluation.anaforaOutput = options.getAnaforaOutput();
+
+ List<Integer> training = trainItems;
+ List<Integer> testing = null;
+ if(options.getTest()){
+ training.addAll(devItems);
+ testing = testItems;
+ }else{
+ testing = devItems;
+ }
+ //do closure on system, but not on gold, to calculate recall
+ evaluation.skipTrain = options.getSkipTrain();
+ evaluation.skipWrite = options.getSkipWrite();
+ if(evaluation.skipTrain && options.getTest()){
+ evaluation.prepareXMIsFor(testing);
+ }else{
+ evaluation.prepareXMIsFor(patientSets);
+ }
+
+ //sort list:
+ Collections.sort(training);
+ Collections.sort(testing);
+
+ //test or train or test
+ evaluation.testOnTrain = options.getTestOnTrain();
+ if(evaluation.testOnTrain){
+ params.stats = evaluation.trainAndTest(training, training);
+ }else{//test on testing set
+ params.stats = evaluation.trainAndTest(training, testing);//training
+ }
+ System.err.println(params.stats);
+
+ if(options.getUseTmp()){
+ // won't work because it's not empty. should we be concerned with this or is it responsibility of
+ // person invoking the tmp flag?
+ FileUtils.deleteRecursive(workingDir);
+ }
+ }catch(ResourceInitializationException e){
+ System.err.println("Error with parameter settings: " + params);
+ e.printStackTrace();
+ }
+ }
+
+ // private ParameterSettings params;
+ protected boolean useClosure;
+ protected boolean useGoldAttributes;
+ protected boolean skipTrain=false;
+ public boolean skipWrite = false;
+ // protected boolean printRelations = false;
+ protected boolean testOnTrain=false;
+
+ public EvaluationOfNeuralJointRelations(
+ File baseDirectory,
+ File rawTextDirectory,
+ File xmlDirectory,
+ XMLFormat xmlFormat,
+ Subcorpus subcorpus,
+ File xmiDirectory,
+ File treebankDirectory,
+ boolean useClosure,
+ boolean printErrors,
+ boolean printRelations,
+ boolean baseline,
+ boolean useGoldAttributes,
+ String kernelParams,
+ ParameterSettings params){
+ super(
+ baseDirectory,
+ rawTextDirectory,
+ xmlDirectory,
+ xmlFormat,
+ subcorpus,
+ xmiDirectory,
+ treebankDirectory,
+ printErrors,
+ printRelations,
+ params);
+ this.params = params;
+ this.useClosure = useClosure;
+ this.printErrors = printErrors;
+ this.printRelations = printRelations;
+ this.useGoldAttributes = useGoldAttributes;
+ this.kernelParams = kernelParams == null ? null : kernelParams.split(" ");
+ }
+
+ // public EvaluationOfTemporalRelations(File baseDirectory, File rawTextDirectory,
+ // File knowtatorXMLDirectory, File xmiDirectory) {
+ //
+ // super(baseDirectory, rawTextDirectory, knowtatorXMLDirectory, xmiDirectory, null);
+ // this.params = defaultParams;
+ // this.printErrors = false;
+ // }
+
+ @Override
+ protected void train(CollectionReader collectionReader, File directory) throws Exception {
+ // if(this.baseline) return;
+ if(this.skipTrain) return;
+ if(!this.skipWrite){
+ AggregateBuilder aggregateBuilder = this.getPreprocessorAggregateBuilder();
+ aggregateBuilder.add(CopyFromGold.getDescription(EventMention.class, TimeMention.class, BinaryTextRelation.class));
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveCrossSentenceRelations.class));
+ if(!this.useGoldAttributes){
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveGoldAttributes.class));
+ }
+ if (this.useClosure) {
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(AddClosure.class));
+ }
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveNonContainsRelations.class));
+ // aggregateBuilder.add(AnalysisEngineFactory.createPrimitiveDescription(AddFlippedOverlap.class));//add flipped overlap instances to training data
+
+ // aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveNonTLINKRelations.class));//remove non tlink relations, such as alinks
+
+ // aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(Overlap2Contains.class));
+
+ // aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(PreserveEventEventRelations.class));
+ // aggregateBuilder.add(AnalysisEngineFactory.createPrimitiveDescription(RemoveNonUMLSEvents.class));
+
+ //add unlabeled nearby system events as potential links:
+ // aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(AddEEPotentialRelations.class));
+ // aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(AddPotentialRelations.class));
+
+ // aggregateBuilder.add(
+ // AnalysisEngineFactory.createEngineDescription(EventEventTokenBasedAnnotator.class,//EventEventTokenBasedAnnotator.class,EventEventPathsBasedAnnotator.class, EventEventTokenAndPosBasedAnnotator, EventEventPathsBasedAnnotator
+ // CleartkAnnotator.PARAM_IS_TRAINING,
+ // true,
+ // DefaultDataWriterFactory.PARAM_DATA_WRITER_CLASS_NAME,
+ // KerasStringOutcomeDataWriter.class,
+ // DirectoryDataWriterFactory.PARAM_OUTPUT_DIRECTORY,
+ // new File(directory,"event-event"),
+ // ScriptStringFeatureDataWriter.PARAM_SCRIPT_DIR,
+ // "scripts/nn/"
+ // ) );
+
+ aggregateBuilder.add(
+ AnalysisEngineFactory.createEngineDescription(JointRelationTokenBasedAnnotator.class,//EventTimeTokenAndPathBasedAnnotator.class,//
+ CleartkAnnotator.PARAM_IS_TRAINING,
+ true,
+ DefaultDataWriterFactory.PARAM_DATA_WRITER_CLASS_NAME,
+ KerasStringOutcomeDataWriter.class,
+ DirectoryDataWriterFactory.PARAM_OUTPUT_DIRECTORY,
+ new File(directory,"joint"),
+ ScriptStringFeatureDataWriter.PARAM_SCRIPT_DIR,
+ "scripts/nn-joint/"
+ ) );
+
+ SimplePipeline.runPipeline(collectionReader, aggregateBuilder.createAggregate());
+ }
+ // HideOutput hider = new HideOutput();
+ JarClassifierBuilder.trainAndPackage(new File(directory,"joint"));
+
+ }
+
+ @Override
+ protected AnnotationStatistics<String> test(CollectionReader collectionReader, File directory)
+ throws Exception {
+ this.useClosure=false;
+ AggregateBuilder aggregateBuilder = this.getPreprocessorAggregateBuilder();
+
+ aggregateBuilder.add(CopyFromGold.getDescription(EventMention.class, TimeMention.class));
+
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(
+ RemoveCrossSentenceRelations.class,
+ RemoveCrossSentenceRelations.PARAM_SENTENCE_VIEW,
+ CAS.NAME_DEFAULT_SOFA,
+ RemoveCrossSentenceRelations.PARAM_RELATION_VIEW,
+ GOLD_VIEW_NAME));
+
+ if (!recallModeEvaluation && this.useClosure) { //closure for gold
+ aggregateBuilder.add(
+ AnalysisEngineFactory.createEngineDescription(AddClosure.class),//AnalysisEngineFactory.createPrimitiveDescription(AddTransitiveContainsRelations.class),
+ CAS.NAME_DEFAULT_SOFA,
+ GOLD_VIEW_NAME);
+ }
+
+ //keep event event tlinks, remove the other relations
+ // aggregateBuilder.add(
+ // AnalysisEngineFactory.createEngineDescription(PreserveEventEventRelations.class),
+ // CAS.NAME_DEFAULT_SOFA,
+ // GOLD_VIEW_NAME);
+
+ //remove non-tlink relations, such as alinks
+ // aggregateBuilder.add(
+ // AnalysisEngineFactory.createEngineDescription(RemoveNonTLINKRelations.class),
+ // CAS.NAME_DEFAULT_SOFA,
+ // GOLD_VIEW_NAME);
+
+ // aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveNonUMLSEvents.class));
+
+
+
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveNonContainsRelations.class),
+ CAS.NAME_DEFAULT_SOFA,
+ GOLD_VIEW_NAME);
+
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveRelations.class));
+ AnalysisEngineDescription aed = null;
+ // aed=AnalysisEngineFactory.createEngineDescription(EventEventTokenBasedAnnotator.class,//EventEventTokenBasedAnnotator.class,EventEventPathsBasedAnnotator.class, EventEventTokenAndPosBasedAnnotator
+ // CleartkAnnotator.PARAM_IS_TRAINING,
+ // false,
+ // GenericJarClassifierFactory.PARAM_CLASSIFIER_JAR_PATH,
+ // new File(new File(directory,"event-event"), "model.jar").getPath());
+ // aed = EventEventRelationAnnotator.createAnnotatorDescription((new File(directory,"event-event/model.jar")).getAbsolutePath());
+ // aggregateBuilder.add(aed);
+ aed = AnalysisEngineFactory.createEngineDescription(JointRelationTokenBasedAnnotator.class,
+ CleartkAnnotator.PARAM_IS_TRAINING,
+ false,
+ GenericJarClassifierFactory.PARAM_CLASSIFIER_JAR_PATH,
+ new File(new File(directory,"joint"), "model.jar").getPath());
+ aggregateBuilder.add(aed);
+
+// aed = DocTimeRelAnnotator.createAnnotatorDescription(new File("target/eval/event-properties/train_and_test/docTimeRel/model.jar").getAbsolutePath());
+// aggregateBuilder.add(aed);
+
+ // aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(CrossSentenceTemporalRelationAnnotator.class));
+ // aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(WithinSentenceBeforeRelationAnnotator.class));
+
+ if(this.anaforaOutput != null){
+ aed = AnalysisEngineFactory.createEngineDescription(WriteAnaforaXML.class, WriteAnaforaXML.PARAM_OUTPUT_DIR, this.anaforaOutput);
+ aggregateBuilder.add(aed, "TimexView", CAS.NAME_DEFAULT_SOFA);
+ }
+
+ File outf = null;
+ if (recallModeEvaluation && this.useClosure) {//add closure for system output
+ aggregateBuilder.add(
+ AnalysisEngineFactory.createEngineDescription(AddClosure.class),//AnalysisEngineFactory.createPrimitiveDescription(AddTransitiveContainsRelations.class),
+ GOLD_VIEW_NAME,
+ CAS.NAME_DEFAULT_SOFA
+ );
+ outf = new File("target/eval/thyme/SystemError_eventEvent_recall_test.txt");
+ }else if (!recallModeEvaluation && this.useClosure){
+ outf = new File("target/eval/thyme/SystemError_eventEvent_precision_test.txt");
+ }else{
+ outf = new File("target/eval/thyme/SystemError_eventEvent_plain_test.txt");
+ }
+
+ PrintWriter outDrop =null;
+// outDrop = new PrintWriter(new BufferedWriter(new FileWriter(outf, false)));
+
+ Function<BinaryTextRelation, ?> getSpan = new Function<BinaryTextRelation, HashableArguments>() {
+ public HashableArguments apply(BinaryTextRelation relation) {
+ return new HashableArguments(relation);
+ }
+ };
+ Function<BinaryTextRelation, String> getOutcome = AnnotationStatistics.annotationToFeatureValue("category");
+
+ AnnotationStatistics<String> stats = new AnnotationStatistics<>();
+ JCasIterator jcasIter =new JCasIterator(collectionReader, aggregateBuilder.createAggregate());
+ JCas jCas = null;
+ while(jcasIter.hasNext()) {
+ jCas = jcasIter.next();
+ JCas goldView = jCas.getView(GOLD_VIEW_NAME);
+ JCas systemView = jCas.getView(CAS.NAME_DEFAULT_SOFA);
+ Collection<BinaryTextRelation> goldRelations = JCasUtil.select(
+ goldView,
+ BinaryTextRelation.class);
+ Collection<BinaryTextRelation> systemRelations = JCasUtil.select(
+ systemView,
+ BinaryTextRelation.class);
+
+ stats.add(goldRelations, systemRelations, getSpan, getOutcome);
+ if(this.printRelations){
+ URI uri = ViewUriUtil.getURI(jCas);
+ String[] path = uri.getPath().split("/");
+ printRelationAnnotations(path[path.length - 1], systemRelations);
+ }
+ if(this.printErrors){
+ Map<HashableArguments, BinaryTextRelation> goldMap = Maps.newHashMap();
+ for (BinaryTextRelation relation : goldRelations) {
+ goldMap.put(new HashableArguments(relation), relation);
+ }
+ Map<HashableArguments, BinaryTextRelation> systemMap = Maps.newHashMap();
+ for (BinaryTextRelation relation : systemRelations) {
+ systemMap.put(new HashableArguments(relation), relation);
+ }
+ Set<HashableArguments> all = Sets.union(goldMap.keySet(), systemMap.keySet());
+ List<HashableArguments> sorted = Lists.newArrayList(all);
+ Collections.sort(sorted);
+ outDrop.println("Doc id: " + ViewUriUtil.getURI(jCas).toString());
+ for (HashableArguments key : sorted) {
+ BinaryTextRelation goldRelation = goldMap.get(key);
+ BinaryTextRelation systemRelation = systemMap.get(key);
+ if (goldRelation == null) {
+ outDrop.println("System added: " + formatRelation(systemRelation));
+ } else if (systemRelation == null) {
+ outDrop.println("System dropped: " + formatRelation(goldRelation));
+ } else if (!systemRelation.getCategory().equals(goldRelation.getCategory())) {
+ String label = systemRelation.getCategory();
+ outDrop.printf("System labeled %s for %s\n", label, formatRelation(goldRelation));
+ } else{
+ outDrop.println("Nailed it! " + formatRelation(systemRelation));
+ }
+ }
+ }
+ }
+// outDrop.close();
+ return stats;
+ }
+
+ public static class RemoveNonUMLSEvents extends org.apache.uima.fit.component.JCasAnnotator_ImplBase {
+ public static final String PARAM_GOLD_VIEW = "GoldView";
+
+ @ConfigurationParameter(name = PARAM_GOLD_VIEW,mandatory=false)
+ private String goldViewName = CAS.NAME_DEFAULT_SOFA;
+
+ @Override
+ public void process(JCas jCas) throws AnalysisEngineProcessException {
+ JCas sysView;
+ JCas goldView;
+ try {
+ sysView = jCas.getView(CAS.NAME_DEFAULT_SOFA);
+ goldView = jCas.getView(PARAM_GOLD_VIEW);
+ } catch (CASException e) {
+ throw new AnalysisEngineProcessException(e);
+ }
+ for(TemporalTextRelation relation : Lists.newArrayList(JCasUtil.select(goldView, TemporalTextRelation.class))){
+ Annotation arg1 = relation.getArg1().getArgument();
+ Annotation arg2 = relation.getArg2().getArgument();
+ boolean arg1Valid = false;
+ boolean arg2Valid = false;
+ for (EventMention event : JCasUtil.selectCovered(sysView, EventMention.class, arg1)){
+ if(!event.getClass().equals(EventMention.class)){
+ arg1Valid = true;
+ break;
+ }
+ }
+ for (EventMention event : JCasUtil.selectCovered(sysView, EventMention.class, arg2)){
+ if(!event.getClass().equals(EventMention.class)){
+ arg2Valid = true;
+ break;
+ }
+ }
+ if(arg1Valid && arg2Valid){
+ // these are the kind we keep.
+ continue;
+ }
+ arg1.removeFromIndexes();
+ arg2.removeFromIndexes();
+ relation.removeFromIndexes();
+ }
+ }
+ }
+
+
+ static void createRelation(JCas jCas, Annotation arg1,
+ Annotation arg2, String category) {
+ RelationArgument relArg1 = new RelationArgument(jCas);
+ relArg1.setArgument(arg1);
+ relArg1.setRole("Arg1");
+ relArg1.addToIndexes();
+ RelationArgument relArg2 = new RelationArgument(jCas);
+ relArg2.setArgument(arg2);
+ relArg2.setRole("Arg2");
+ relArg2.addToIndexes();
+ TemporalTextRelation relation = new TemporalTextRelation(jCas);
+ relation.setArg1(relArg1);
+ relation.setArg2(relArg2);
+ relation.setCategory(category);
+ relation.addToIndexes();
+
+ }
+
+
+ /*
+ private static String formatRelation(BinaryTextRelation relation) {
+ IdentifiedAnnotation arg1 = (IdentifiedAnnotation)relation.getArg1().getArgument();
+ IdentifiedAnnotation arg2 = (IdentifiedAnnotation)relation.getArg2().getArgument();
+ String text = arg1.getCAS().getDocumentText();
+ int begin = Math.min(arg1.getBegin(), arg2.getBegin());
+ int end = Math.max(arg1.getBegin(), arg2.getBegin());
+ begin = Math.max(0, begin - 50);
+ end = Math.min(text.length(), end + 50);
+ return String.format(
+ "%s(%s(type=%d), %s(type=%d)) in ...%s...",
+ relation.getCategory(),
+ arg1.getCoveredText(),
+ arg1.getTypeID(),
+ arg2.getCoveredText(),
+ arg2.getTypeID(),
+ text.substring(begin, end).replaceAll("[\r\n]", " "));
+ }
+
+ private static void printRelationAnnotations(String fileName, Collection<BinaryTextRelation> relations) {
+
+ for(BinaryTextRelation binaryTextRelation : relations) {
+
+ Annotation arg1 = binaryTextRelation.getArg1().getArgument();
+ Annotation arg2 = binaryTextRelation.getArg2().getArgument();
+
+ String arg1Type = arg1.getClass().getSimpleName();
+ String arg2Type = arg2.getClass().getSimpleName();
+
+ int arg1Begin = arg1.getBegin();
+ int arg1End = arg1.getEnd();
+ int arg2Begin = arg2.getBegin();
+ int arg2End = arg2.getEnd();
+
+ String category = binaryTextRelation.getCategory();
+
+ System.out.format("%s\t%s\t%s\t%d\t%d\t%s\t%d\t%d\n",
+ fileName, category, arg1Type, arg1Begin, arg1End, arg2Type, arg2Begin, arg2End);
+ }
+ }
+ */
+
+
+
+ // @SuppressWarnings("unchecked")
+ // private static <SPAN> Collection<BinaryTextRelation> getDuplicateRelations(
+ // Collection<BinaryTextRelation> goldRelations,
+ // Function<BinaryTextRelation, ?> getSpan) {
+ // Set<BinaryTextRelation> duplicateRelations = Sets.newHashSet();
+ // //build a multimap that map gold span to gold relation
+ // Multimap<SPAN, BinaryTextRelation> spanToRelation = HashMultimap.create();
+ // for (BinaryTextRelation relation : goldRelations) {
+ // spanToRelation.put((SPAN) getSpan.apply(relation), relation);
+ // }
+ // for (SPAN span: spanToRelation.keySet()){
+ // Collection<BinaryTextRelation> relations = spanToRelation.get(span);
+ // if(relations.size()>1){//if same span maps to multiple relations
+ // duplicateRelations.addAll(relations);
+ // }
+ // }
+ // return duplicateRelations;
+ // }
+
+ // private static Collection<BinaryTextRelation> removeNonGoldRelations(
+ // Collection<BinaryTextRelation> systemRelations, Collection<BinaryTextRelation> goldRelations) {
+ // //remove non-gold pairs from system relations:
+ // Set<BinaryTextRelation> goodSys = Sets.newHashSet();
+ //
+ // for(BinaryTextRelation sysrel : systemRelations){
+ // Annotation sysArg1 = sysrel.getArg1().getArgument();
+ // Annotation sysArg2 = sysrel.getArg2().getArgument();
+ // for(BinaryTextRelation goldrel : goldRelations){
+ // Annotation goldArg1 = goldrel.getArg1().getArgument();
+ // Annotation goldArg2 = goldrel.getArg2().getArgument();
+ // if(matchSpan(sysArg1, goldArg1) && matchSpan(sysArg2, goldArg2)){
+ // goodSys.add(sysrel);
+ // continue;
+ // }else if (matchSpan(sysArg2, goldArg1) && matchSpan(sysArg1, goldArg2)){//the order of system pair was flipped
+ // if(sysrel.getCategory().equals("OVERLAP")){ //if the relation is overlap, and the arg order was flipped, then change back the order
+ // RelationArgument tempArg = (RelationArgument) sysrel.getArg1().clone();
+ // sysrel.setArg1((RelationArgument) sysrel.getArg2().clone());
+ // sysrel.setArg2(tempArg);
+ // }//for other types of relation, still maintain the type.
+ // goodSys.add(sysrel);
+ // continue;
+ // }
+ // }
+ // }
+ //
+ // return goodSys;
+ // }
+
+
+
+
+ public static class RemoveNonTLINKRelations extends JCasAnnotator_ImplBase {
+ @Override
+ public void process(JCas jCas) throws AnalysisEngineProcessException {
+ for (BinaryTextRelation relation : Lists.newArrayList(JCasUtil.select(
+ jCas,
+ BinaryTextRelation.class))) {
+ if (!(relation instanceof TemporalTextRelation)) {
+ relation.getArg1().removeFromIndexes();
+ relation.getArg2().removeFromIndexes();
+ relation.removeFromIndexes();
+ }
+ }
+ }
+ }
+
+ public static class RemoveCrossSentenceRelations extends org.apache.uima.fit.component.JCasAnnotator_ImplBase {
+
+ public static final String PARAM_SENTENCE_VIEW = "SentenceView";
+
+ @ConfigurationParameter(name = PARAM_SENTENCE_VIEW,mandatory=false)
+ private String sentenceViewName = CAS.NAME_DEFAULT_SOFA;
+
+ public static final String PARAM_RELATION_VIEW = "RelationView";
+
+ @ConfigurationParameter(name = PARAM_RELATION_VIEW,mandatory=false)
+ private String relationViewName = CAS.NAME_DEFAULT_SOFA;
+
+ @Override
+ public void process(JCas jCas) throws AnalysisEngineProcessException {
+ JCas sentenceView, relationView;
+ try {
+ sentenceView = jCas.getView(this.sentenceViewName);
+ relationView = jCas.getView(this.relationViewName);
+ } catch (CASException e) {
+ throw new AnalysisEngineProcessException(e);
+ }
+
+ // map events and times to the sentences that contain them
+ Map<IdentifiedAnnotation, Integer> sentenceIndex = Maps.newHashMap();
+ int index = -1;
+ for (Sentence sentence : JCasUtil.select(sentenceView, Sentence.class)) {
+ ++index;
+ for (EventMention event : JCasUtil.selectCovered(relationView, EventMention.class, sentence)) {
+ sentenceIndex.put(event, index);
+ }
+ for (TimeMention time : JCasUtil.selectCovered(relationView, TimeMention.class, sentence)) {
+ sentenceIndex.put(time, index);
+ }
+ }
+
+ // remove any relations that are in different sentences.
+ for (BinaryTextRelation relation : Lists.newArrayList(JCasUtil.select(
+ relationView,
+ BinaryTextRelation.class))) {
+ Integer sent1 = sentenceIndex.get(relation.getArg1().getArgument());
+ Integer sent2 = sentenceIndex.get(relation.getArg2().getArgument());
+ if (sent1 == null || sent2 == null || !sent1.equals(sent2)) {
+ relation.getArg1().removeFromIndexes();
+ relation.getArg2().removeFromIndexes();
+ relation.removeFromIndexes();
+ }
+ }
+ }
+ }
+
+
+ public static class RemoveRelations extends org.apache.uima.fit.component.JCasAnnotator_ImplBase {
+ @Override
+ public void process(JCas jCas) throws AnalysisEngineProcessException {
+ for (BinaryTextRelation relation : Lists.newArrayList(JCasUtil.select(
+ jCas,
+ BinaryTextRelation.class))) {
+ relation.getArg1().removeFromIndexes();
+ relation.getArg2().removeFromIndexes();
+ relation.removeFromIndexes();
+ }
+ }
+ }
+
+
+ public static class AddTransitiveContainsRelations extends org.apache.uima.fit.component.JCasAnnotator_ImplBase {
+
+ @Override
+ public void process(JCas jCas) throws AnalysisEngineProcessException {
+
+ // collect many-to-many mappings of containment relations
+ Multimap<Annotation, Annotation> isContainedIn = HashMultimap.create();
+ Multimap<Annotation, Annotation> contains = HashMultimap.create();
+ Set<BinaryTextRelation> containsRelations = Sets.newHashSet();
+ for (BinaryTextRelation relation : JCasUtil.select(jCas, BinaryTextRelation.class)) {
+ if (relation.getCategory().equals("CONTAINS")) {
+ containsRelations.add(relation);
+ Annotation arg1 = relation.getArg1().getArgument();
+ Annotation arg2 = relation.getArg2().getArgument();
+ contains.put(arg1, arg2);
+ isContainedIn.put(arg2, arg1);
+ }
+ }
+
+ // look for X -> Y -> Z containment chains and add X -> Z relations
+ Deque<Annotation> todo = new ArrayDeque<>(isContainedIn.keySet());
+ while (!todo.isEmpty()) {
+ Annotation next = todo.removeFirst();
+ for (Annotation parent : Lists.newArrayList(isContainedIn.get(next))) {
+ for (Annotation grandParent : Lists.newArrayList(isContainedIn.get(parent))) {
+ if (!isContainedIn.containsEntry(next, grandParent)) {
+ isContainedIn.put(next, grandParent);
+ contains.put(grandParent, next);
+
+ // once X -> Z has been added, we need to re-do all W where W -> X
+ for (Annotation child : contains.get(next)) {
+ todo.add(child);
+ }
+ }
+ }
+ }
+ }
+
+ // remove old relations
+ for (BinaryTextRelation relation : containsRelations) {
+ relation.getArg1().removeFromIndexes();
+ relation.getArg2().removeFromIndexes();
+ relation.removeFromIndexes();
+ }
+
+ // add new, transitive relations
+ for (Annotation contained : isContainedIn.keySet()) {
+ for (Annotation container : isContainedIn.get(contained)) {
+ RelationArgument arg1 = new RelationArgument(jCas);
+ arg1.setArgument(container);
+ RelationArgument arg2 = new RelationArgument(jCas);
+ arg2.setArgument(contained);
+ BinaryTextRelation relation = new BinaryTextRelation(jCas);
+ relation.setArg1(arg1);
+ relation.setArg2(arg2);
+ relation.setCategory("CONTAINS");
+ arg1.addToIndexes();
+ arg2.addToIndexes();
+ relation.addToIndexes();
+ }
+ }
+ }
+
+ }
+
+ public static class AddContain2Overlap extends org.apache.uima.fit.component.JCasAnnotator_ImplBase {
+
+ @Override
+ public void process(JCas jCas) throws AnalysisEngineProcessException {
+
+ Set<BinaryTextRelation> containsRelations = Sets.newHashSet();
+ for (BinaryTextRelation relation : JCasUtil.select(jCas, BinaryTextRelation.class)) {
+ if (relation.getCategory().equals("CONTAINS")) {
+ containsRelations.add(relation);
+ }
+ }
+
+ for (BinaryTextRelation relation : containsRelations) {
+ RelationArgument arg1 = (RelationArgument) relation.getArg1().clone();
+ RelationArgument arg2 = (RelationArgument) relation.getArg2().clone();
+ BinaryTextRelation newrelation = new BinaryTextRelation(jCas);
+ newrelation.setArg1(arg1);
+ newrelation.setArg2(arg2);
+ newrelation.setCategory("OVERLAP");
+ arg1.addToIndexes();
+ arg2.addToIndexes();
+ newrelation.addToIndexes();
+ }
+ }
+ }
+
+ public static class AddFlippedOverlap extends org.apache.uima.fit.component.JCasAnnotator_ImplBase {
+
+ @Override
+ public void process(JCas jCas) throws AnalysisEngineProcessException {
+
+ Set<BinaryTextRelation> overlapRelations = Sets.newHashSet();
+ Multimap<Annotation, Annotation> overlaps = HashMultimap.create();
+ for (BinaryTextRelation relation : JCasUtil.select(jCas, BinaryTextRelation.class)) {
+ if (relation.getCategory().equals("OVERLAP")) {
+ overlapRelations.add(relation);
+ Annotation arg1 = relation.getArg1().getArgument();
+ Annotation arg2 = relation.getArg2().getArgument();
+ overlaps.put(arg1, arg2);
+ }
+ }
+
+ for (BinaryTextRelation orelation : overlapRelations) {
+ Annotation argA = orelation.getArg1().getArgument();
+ Annotation argB = orelation.getArg2().getArgument();
+ //add overlap
+ if (!overlaps.containsEntry(argB, argA)) {
+ //create a new flipped relation:
+ RelationArgument arg1 = new RelationArgument(jCas);
+ arg1.setArgument(argB);
+ RelationArgument arg2 = new RelationArgument(jCas);
+ arg2.setArgument(argA);
+ BinaryTextRelation relation = new BinaryTextRelation(jCas);
+ relation.setArg1(arg1);
+ relation.setArg2(arg2);
+ relation.setCategory("OVERLAP");
+ arg1.addToIndexes();
+ arg2.addToIndexes();
+ relation.addToIndexes();
+ overlaps.put(argB, argA);
+ }
+
+ }
+ }
+ }
+
+ public static class AddClosure extends JCasAnnotator_ImplBase {
+
+ @Override
+ public void process(JCas jCas) throws AnalysisEngineProcessException {
+
+ Multimap<List<Annotation>, BinaryTextRelation> annotationsToRelation = HashMultimap.create();
+ for (BinaryTextRelation relation : JCasUtil.select(jCas, BinaryTextRelation.class)){
+ String relationType = relation.getCategory();
+ if(validTemporalType(relationType)){
+ Annotation arg1 = relation.getArg1().getArgument();
+ Annotation arg2 = relation.getArg2().getArgument();
+ annotationsToRelation.put(Arrays.asList(arg1, arg2), relation);
+ }
+ }
+ for (List<Annotation> span: Lists.newArrayList(annotationsToRelation.keySet())){
+ Collection<BinaryTextRelation> relations = annotationsToRelation.get(span);
+ if(relations.size()>1){//if same span maps to multiple relations
+ Set<String> types = Sets.newHashSet();
+ for(BinaryTextRelation relation: relations){
+ types.add(relation.getCategory());
+ }
+ if(types.size()>1){
+ for(BinaryTextRelation relation: Lists.newArrayList(relations)){
+ annotationsToRelation.remove(span, relation);
+ relation.getArg1().removeFromIndexes();
+ relation.getArg2().removeFromIndexes();
+ relation.removeFromIndexes();
+ }
+ }else if(types.size()==1){
+ for (int i =1; i< relations.size(); i++){
+ BinaryTextRelation relation = (BinaryTextRelation) relations.toArray()[i];
+ annotationsToRelation.remove(span, relation);
+ relation.getArg1().removeFromIndexes();
+ relation.getArg2().removeFromIndexes();
+ relation.removeFromIndexes();
+ }
+ }
+ }
+ }
+
+ ArrayList<BinaryTextRelation> temporalRelation = new ArrayList<>(annotationsToRelation.values());//new ArrayList<BinaryTextRelation>();
+ // Map<List<Annotation>, BinaryTextRelation> temporalRelationLookup = new HashMap<List<Annotation>, BinaryTextRelation>();
+ //
+ // for (BinaryTextRelation relation : JCasUtil.select(jCas, BinaryTextRelation.class)){
+ // String relationType = relation.getCategory();
+ // if(validTemporalType(relationType)){
+ // Annotation arg1 = relation.getArg1().getArgument();
+ // Annotation arg2 = relation.getArg2().getArgument();
+ // BinaryTextRelation tempRelation = temporalRelationLookup.get(Arrays.asList(arg1, arg2));
+ // if( tempRelation == null){
+ // temporalRelation.add(relation);
+ // temporalRelationLookup.put(Arrays.asList(arg1, arg2), relation);
+ // }else{//if there is duplicate
+ // relation.getArg1().removeFromIndexes();
+ // relation.getArg2().removeFromIndexes();
+ // relation.removeFromIndexes();
+ // }
+ //
+ // }
+ // }
+
+ if (!temporalRelation.isEmpty()){
+ TLinkTypeArray2 relationArray = new TLinkTypeArray2(temporalRelation, new AnnotationIdCollection(temporalRelation));
+
+ int addedCount = 0;
+ for (BinaryTextRelation relation : relationArray.getClosedTlinks(jCas)) {
+ RelationArgument arg1 = relation.getArg1();
+ RelationArgument arg2 = relation.getArg2();
+ String relationType = relation.getCategory();
+ if(relationType.equals("CONTAINED-BY")||relationType.equals("AFTER")){//ignore these two categories, because their reciprocal already exist.
+ continue;
+ }
+ //check if the inferred relation new:
+ Collection<BinaryTextRelation> relations = annotationsToRelation.get(Arrays.asList(arg1.getArgument(), arg2.getArgument()));
+ if(relations.isEmpty()){ //if haven't seen this inferred relation before, then add this relation
+ arg1.addToIndexes();
+ arg2.addToIndexes();
+ relation.addToIndexes();
+ addedCount++;
+ }
+ }
+
+ System.out.println( "**************************************************************");
+ System.out.println( "Finally added closure relations: " + addedCount );
+ System.out.println( "**************************************************************");
+ }
+
+ }
+
+ private static boolean validTemporalType(String relationType) {
+ if(relationType.equals("CONTAINS")||relationType.equals("OVERLAP")||relationType.equals("BEFORE")||relationType.equals("ENDS-ON")||relationType.equals("BEGINS-ON"))
+ return true;
+ return false;
+ }
+ }
+
+ // public static class AddTransitiveBeforeAndOnRelations extends JCasAnnotator_ImplBase {
+ //
+ // @Override
+ // public void process(JCas jCas) throws AnalysisEngineProcessException {
+ //
+ // // collect many-to-many mappings of containment relations
+ // Multimap<Annotation, Annotation> contains = HashMultimap.create();
+ // Multimap<Annotation, Annotation> before = HashMultimap.create();
+ // Multimap<Annotation, Annotation> endson = HashMultimap.create();
+ // Multimap<Annotation, Annotation> beginson = HashMultimap.create();
+ // Set<BinaryTextRelation> beforeRel = Sets.newHashSet();
+ //
+ // for (BinaryTextRelation relation : JCasUtil.select(jCas, BinaryTextRelation.class)) {
+ // if (relation.getCategory().equals("CONTAINS")) {
+ // Annotation arg1 = relation.getArg1().getArgument();
+ // Annotation arg2 = relation.getArg2().getArgument();
+ // contains.put(arg1, arg2);
+ // }else if (relation.getCategory().equals("BEFORE")) {
+ // Annotation arg1 = relation.getArg1().getArgument();
+ // Annotation arg2 = relation.getArg2().getArgument();
+ // before.put(arg1, arg2);
+ // beforeRel.add(relation);
+ // }else if (relation.getCategory().equals("ENDS-ON")) {
+ // Annotation arg1 = relation.getArg1().getArgument();
+ // Annotation arg2 = relation.getArg2().getArgument();
+ // endson.put(arg1, arg2);
+ // if (!endson.containsEntry(arg2, arg1)) {
+ // endson.put(arg2, arg1);
+ // }
+ // }else if (relation.getCategory().equals("BEGINS-ON")) {
+ // Annotation arg1 = relation.getArg1().getArgument();
+ // Annotation arg2 = relation.getArg2().getArgument();
+ // beginson.put(arg1, arg2);
+ // if (!beginson.containsEntry(arg2, arg1)) {
+ // beginson.put(arg2, arg1);
+ // }
+ // }
+ // }
+ //
+ // // for A BEFORE B, check if A and B Contain anything
+ // for (BinaryTextRelation brelation : beforeRel) {
+ // Annotation argA = brelation.getArg1().getArgument();
+ // Annotation argB = brelation.getArg2().getArgument();
+ // //add contained before
+ // for (Annotation childA : contains.get(argA)) {
+ // for (Annotation childB : contains.get(argB)) {
+ // if (!before.containsEntry(childA, childB)) {
+ // //create a new before relation:
+ // RelationArgument arg1 = new RelationArgument(jCas);
+ // arg1.setArgument(childA);
+ // RelationArgument arg2 = new RelationArgument(jCas);
+ // arg2.setArgument(childB);
+ // BinaryTextRelation relation = new BinaryTextRelation(jCas);
+ // relation.setArg1(arg1);
+ // relation.setArg2(arg2);
+ // relation.setCategory("BEFORE");
+ // arg1.addToIndexes();
+ // arg2.addToIndexes();
+ // relation.addToIndexes();
+ // before.put(childA, childB);
+ // }
+ // }
+ // }
+ // //add ends-on A
+ // for (Annotation endsOnA : endson.get(argA)) {
+ // if (!before.containsEntry(endsOnA, argB)) {
+ // //create a new before relation:
+ // RelationArgument arg1 = new RelationArgument(jCas);
+ // arg1.setArgument(endsOnA);
+ // RelationArgument arg2 = new RelationArgument(jCas);
+ // arg2.setArgument(argB);
+ // BinaryTextRelation relation = new BinaryTextRelation(jCas);
+ // relation.setArg1(arg1);
+ // relation.setArg2(arg2);
+ // relation.setCategory("BEFORE");
+ // arg1.addToIndexes();
+ // arg2.addToIndexes();
+ // relation.addToIndexes();
+ // before.put(endsOnA, argB);
+ // }
+ // }
+ // //add begins-on B
+ // for (Annotation beginsOnB : beginson.get(argB)) {
+ // if (!before.containsEntry(argA, beginsOnB)) {
+ // //create a new before relation:
+ // RelationArgument arg1 = new RelationArgument(jCas);
+ // arg1.setArgument(argA);
+ // RelationArgument arg2 = new RelationArgument(jCas);
+ // arg2.setArgument(beginsOnB);
+ // BinaryTextRelation relation = new BinaryTextRelation(jCas);
+ // relation.setArg1(arg1);
+ // relation.setArg2(arg2);
+ // relation.setCategory("BEFORE");
+ // arg1.addToIndexes();
+ // arg2.addToIndexes();
+ // relation.addToIndexes();
+ // before.put(argA, beginsOnB);
+ // }
+ // }
+ // }
+ // }
+ //
+ // }
+}