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 2016/07/19 17:14:36 UTC
svn commit: r1753407 [3/3] - in
/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal: ae/
ae/feature/ eval/ keras/
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/eval/NeuralEventTimeRelationsEvaluation.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/eval/NeuralEventTimeRelationsEvaluation.java?rev=1753407&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/eval/NeuralEventTimeRelationsEvaluation.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/eval/NeuralEventTimeRelationsEvaluation.java Tue Jul 19 17:14:35 2016
@@ -0,0 +1,1390 @@
+/*
+ * 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.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.HashMap;
+//import java.util.HashSet;
+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.EventTimeSelfRelationAnnotator;
+import org.apache.ctakes.temporal.ae.NeuralEventTimeSelfRelationAnnotator;
+import org.apache.ctakes.temporal.ae.TemporalRelationExtractorAnnotator;
+//import org.apache.ctakes.temporal.ae.EventTimeSyntacticAnnotator;
+//import org.apache.ctakes.temporal.ae.EventTimeRelationAnnotator;
+//import org.apache.ctakes.temporal.ae.EventEventRelationAnnotator;
+import org.apache.ctakes.temporal.ae.baselines.RecallBaselineEventTimeRelationAnnotator;
+import org.apache.ctakes.temporal.eval.EvaluationOfEventTimeRelations.ParameterSettings;
+//import org.apache.ctakes.temporal.ae.feature.selection.ZscoreNormalizationExtractor; //for normalization
+import org.apache.ctakes.temporal.eval.EvaluationOfTemporalRelations_ImplBase.RemoveGoldAttributes;
+import org.apache.ctakes.temporal.eval.EvaluationOfTemporalRelations_ImplBase.RemoveNonContainsRelations;
+import org.apache.ctakes.temporal.eval.Evaluation_ImplBase.WriteAnaforaXML;
+import org.apache.ctakes.temporal.keras.KerasStringOutcomeDataWriter;
+import org.apache.ctakes.temporal.keras.ScriptStringOutcomeDataWriter;
+//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.syntax.WordToken;
+//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.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.Instance; //for normalization
+//import org.cleartk.ml.feature.transform.InstanceDataWriter;//for normalization
+//import org.cleartk.ml.feature.transform.InstanceStream;//for normalization
+import org.cleartk.ml.jar.JarClassifierBuilder;
+import org.cleartk.ml.liblinear.LibLinearStringOutcomeDataWriter;
+import org.cleartk.ml.libsvm.LibSvmStringOutcomeDataWriter;
+//import org.cleartk.ml.tksvmlight.TkSvmLightStringOutcomeDataWriter;
+import org.cleartk.ml.tksvmlight.model.CompositeKernel;
+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;
+
+public class NeuralEventTimeRelationsEvaluation 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 getTestOnTrain();
+
+ @Option
+ public boolean getSkipWrite();
+ }
+
+ // 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;
+
+ static int sysRelationCount;
+ static int closeRelationCount;
+ static int goldRelationCount;
+ static int closeGoldRelationCount;
+
+ public static void main(String[] args) throws Exception {
+ sysRelationCount = 0;
+ closeRelationCount = 0;
+ goldRelationCount = 0;
+ closeGoldRelationCount = 0;
+
+ 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("target/eval/thyme/");
+ if(!workingDir.exists()) workingDir.mkdirs();
+ if(options.getUseTmp()){
+ File tempModelDir = File.createTempFile("temporal", null, workingDir);
+ tempModelDir.delete();
+ tempModelDir.mkdir();
+ workingDir = tempModelDir;
+ }
+ NeuralEventTimeRelationsEvaluation evaluation = new NeuralEventTimeRelationsEvaluation(
+ 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);
+
+ if(options.getI2B2Output()!=null) evaluation.setI2B2Output(options.getI2B2Output() + "/temporal-relations/event-time");
+ 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);
+ }
+
+ // evaluation.printErrors = true;
+
+ //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(options.getKernelParams() == null ? params : options.getKernelParams());
+ // System.err.println("No closure on gold::Closure on System::Recall Mode");
+ System.err.println(params.stats);
+
+ System.err.println("System predict relations #: "+ sysRelationCount);
+ System.err.println("# of system relations whose arguments are close: "+ closeRelationCount);
+ System.err.println("Gold relations #: "+ goldRelationCount);
+ System.err.println("# of gold relations whose arguments are close: "+ closeGoldRelationCount);
+
+ 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;
+ private boolean baseline;
+ protected boolean useClosure;
+ protected boolean useGoldAttributes;
+ protected boolean skipTrain=false;
+ public boolean skipWrite = false;
+ protected boolean testOnTrain=false;
+ // protected boolean printRelations = false;
+
+ public NeuralEventTimeRelationsEvaluation(
+ 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.baseline = baseline;
+ 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.createPrimitiveDescription(AddTransitiveContainsRelations.class));
+ // aggregateBuilder.add(AnalysisEngineFactory.createPrimitiveDescription(AddContain2Overlap.class));
+ // aggregateBuilder.add(AnalysisEngineFactory.createPrimitiveDescription(AddTransitiveBeforeAndOnRelations.class));
+ }
+// aggregateBuilder.add(AnalysisEngineFactory.createPrimitiveDescription(RemoveNonContainsRelations.class));
+ // aggregateBuilder.add(AnalysisEngineFactory.createPrimitiveDescription(AddFlippedOverlap.class));//add flipped overlap instances to training data
+
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(Overlap2Contains.class));
+
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveEventEventRelations.class));
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveNonContainsRelations.class));
+
+ //count how many sentences have timex, and how many sentences have only one timex
+ //aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(CountSentenceContainsTimes.class));
+
+ //add unlabeled nearby system events as potential links:
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(AddPotentialRelations.class));
+
+ //directory = new File(directory,"event-time");//for normalization
+
+ aggregateBuilder.add(
+ AnalysisEngineFactory.createEngineDescription(NeuralEventTimeSelfRelationAnnotator.class,
+ CleartkAnnotator.PARAM_IS_TRAINING,
+ true,
+ DefaultDataWriterFactory.PARAM_DATA_WRITER_CLASS_NAME,
+ KerasStringOutcomeDataWriter.class,
+ DirectoryDataWriterFactory.PARAM_OUTPUT_DIRECTORY,
+ new File(directory,"event-time"),
+ ScriptStringOutcomeDataWriter.PARAM_SCRIPT_DIR,
+ "scripts/keras/"
+ ) );
+
+ // EventTimeSelfRelationAnnotator.createDataWriterDescription(
+ // LibLinearStringOutcomeDataWriter.class,
+ // new File(directory,"event-time"),
+ // params.probabilityOfKeepingANegativeExample));
+ SimplePipeline.runPipeline(collectionReader, aggregateBuilder.createAggregate());
+
+
+ /**
+ //normalize features:
+ Iterable<Instance<String>> instances = InstanceStream.loadFromDirectory(directory);
+ // Collect MinMax stats for feature normalization
+ ZscoreNormalizationExtractor<String, Annotation> featureTransformExtractor = TemporalRelationExtractorAnnotator.createMinMaxNormalizationExtractor();
+ featureTransformExtractor.train(instances);
+ featureTransformExtractor.save(TemporalRelationExtractorAnnotator.createMinMaxNormalizationExtractorURI(directory));
+ // now write in the libsvm format
+ LibLinearStringOutcomeDataWriter dataWriter = new LibLinearStringOutcomeDataWriter(directory);
+ for (Instance<String> instance : instances) {
+ dataWriter.write(featureTransformExtractor.transform(instance));
+ }
+ dataWriter.finish();
+ */
+ String[] optArray;
+
+ if(this.kernelParams == null){
+ ArrayList<String> svmOptions = new ArrayList<>();
+ svmOptions.add("-c"); svmOptions.add(""+params.svmCost); // svm cost
+ svmOptions.add("-t"); svmOptions.add(""+params.svmKernelIndex); // kernel index
+ svmOptions.add("-d"); svmOptions.add("3"); // degree parameter for polynomial
+ svmOptions.add("-g"); svmOptions.add(""+params.svmGamma);
+ if(params.svmKernelIndex==ParameterSettings.SVM_KERNELS.indexOf("tk")){
+ svmOptions.add("-S"); svmOptions.add(""+params.secondKernelIndex); // second kernel index (similar to -t) for composite kernel
+ String comboFlag = (params.comboOperator == ComboOperator.SUM ? "+" : params.comboOperator == ComboOperator.PRODUCT ? "*" : params.comboOperator == ComboOperator.TREE_ONLY ? "T" : "V");
+ svmOptions.add("-C"); svmOptions.add(comboFlag);
+ svmOptions.add("-L"); svmOptions.add(""+params.lambda);
+ svmOptions.add("-T"); svmOptions.add(""+params.tkWeight);
+ svmOptions.add("-N"); svmOptions.add("3"); // normalize trees and features
+ }
+ optArray = svmOptions.toArray(new String[]{});
+ }else{
+ optArray = this.kernelParams;
+ for(int i = 0; i < optArray.length; i+=2){
+ optArray[i] = "-" + optArray[i];
+ }
+ }
+
+ //calculate class-wise weights:
+ // String[] weightArray=new String[TemporalRelationExtractorAnnotator.category_frequency.size()*2+4];
+ // int weight_idx = 0;
+ float baseFreq = TemporalRelationExtractorAnnotator.category_frequency.get(TemporalRelationExtractorAnnotator.NO_RELATION_CATEGORY);
+ for( Map.Entry<String, Integer> entry: TemporalRelationExtractorAnnotator.category_frequency.entrySet()){
+ // weightArray[weight_idx*2] = "-w"+Integer.toString(weight_idx + 1);
+ float weight = baseFreq/entry.getValue();
+ // weightArray[weight_idx*2+1] = Float.toString(weight);
+ // weight_idx ++;
+ System.err.println("Category:"+entry.getKey()+" freq:"+entry.getValue() + " weight:"+weight);
+ }
+
+ }
+ // weightArray[weight_idx*2] = "-c";
+ // weightArray[weight_idx*2+1] = optArray[1];
+ // weight_idx ++;
+ // weightArray[weight_idx*2] = "-B";
+ // weightArray[weight_idx*2+1] = optArray[3];
+ // HideOutput hider = new HideOutput();
+ JarClassifierBuilder.trainAndPackage(new File(directory,"event-time"));//, weightArray
+ }
+
+ @Override
+ protected AnnotationStatistics<String> test(CollectionReader collectionReader, File directory)
+ throws Exception {
+ this.useClosure=false;//don't do closure for test
+ 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);
+ }
+
+ aggregateBuilder.add(
+ AnalysisEngineFactory.createEngineDescription(RemoveEventEventRelations.class),
+ CAS.NAME_DEFAULT_SOFA,
+ GOLD_VIEW_NAME);
+
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveNonContainsRelations.class),
+ CAS.NAME_DEFAULT_SOFA,
+ GOLD_VIEW_NAME);
+ // aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveNonUMLSEtEvents.class));
+
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(RemoveRelations.class));
+ aggregateBuilder.add(this.baseline ? RecallBaselineEventTimeRelationAnnotator.createAnnotatorDescription(directory) :
+ // EventTimeSelfRelationAnnotator.createEngineDescription(new File(directory,"event-time")));
+ AnalysisEngineFactory.createEngineDescription(NeuralEventTimeSelfRelationAnnotator.class,
+ CleartkAnnotator.PARAM_IS_TRAINING,
+ false,
+ GenericJarClassifierFactory.PARAM_CLASSIFIER_JAR_PATH,
+ new File(new File(directory,"event-time"), "model.jar").getPath()));
+
+ //count how many system predicted relations, their arguments are close to each other, without any other event in between
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(CountCloseRelation.class));
+
+ if(this.anaforaOutput != null){
+ aggregateBuilder.add(AnalysisEngineFactory.createEngineDescription(WriteAnaforaXML.class, WriteAnaforaXML.PARAM_OUTPUT_DIR, this.anaforaOutput), "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
+ );
+ aggregateBuilder.add(
+ AnalysisEngineFactory.createEngineDescription(RemoveEventEventRelations.class),
+ GOLD_VIEW_NAME,
+ CAS.NAME_DEFAULT_SOFA
+ );
+ outf = new File("target/eval/thyme/SystemError_eventTime_recall_test.txt");
+ }else if (!recallModeEvaluation && this.useClosure){
+ outf = new File("target/eval/thyme/SystemError_eventTime_precision_test.txt");
+ }else{
+ outf = new File("target/eval/thyme/SystemError_eventTime_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);
+
+ //newly add
+ // systemRelations = removeNonGoldRelations(systemRelations, goldRelations, getSpan);//for removing non-gold pairs
+ // systemRelations = correctArgOrder(systemRelations, goldRelations);//change the argument order of "OVERLAP" relation, if the order is flipped
+ //find duplicates in gold relations:
+ // Collection<BinaryTextRelation> duplicateGoldRelations = getDuplicateRelations(goldRelations, getSpan);
+ // if(!duplicateGoldRelations.isEmpty()){
+ // System.err.println("******Duplicate gold relations in : " + ViewURIUtil.getURI(jCas).toString());
+ // for (BinaryTextRelation rel : duplicateGoldRelations){
+ // System.err.println("Duplicate : "+ formatRelation(rel));
+ // }
+ // }
+ //end newly add
+
+ 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 RemoveNonUMLSEtEvents 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();
+ Annotation argEvent = null;
+ if(arg1 instanceof EventMention && arg2 instanceof TimeMention){
+ argEvent = arg1;
+ }else if(arg1 instanceof TimeMention && arg2 instanceof EventMention){
+ argEvent = arg2;
+ }else{
+ continue;
+ }
+ boolean eventValid = false;
+ for (EventMention event : JCasUtil.selectCovered(sysView, EventMention.class, argEvent)){
+ if(!event.getClass().equals(EventMention.class)){
+ eventValid = true;
+ break;
+ }
+ }
+ if(eventValid){
+ // these are the kind we keep.
+ continue;
+ }
+ arg1.removeFromIndexes();
+ arg2.removeFromIndexes();
+ relation.removeFromIndexes();
+ }
+ }
+ }
+
+ /*
+ 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);
+ }
+ }
+ */
+
+ // private static <SPAN_TYPE> Collection<BinaryTextRelation> removeNonGoldRelations(
+ // Collection<BinaryTextRelation> systemRelations,
+ // Collection<BinaryTextRelation> goldRelations, Function<BinaryTextRelation, ?> getSpan) {
+ // //remove non-gold pairs from system relations:
+ // Set<BinaryTextRelation> goodSys = Sets.newHashSet();
+ // Set<SPAN_TYPE> goldspans = new HashSet<>();
+ //
+ // for (BinaryTextRelation relation : goldRelations) {
+ // goldspans.add(((SPAN_TYPE) getSpan.apply(relation)));
+ // }
+ //
+ // for (BinaryTextRelation relation : systemRelations) {
+ // if (goldspans.contains(((SPAN_TYPE) getSpan.apply(relation)))) {
+ // goodSys.add(relation);
+ // }
+ // }
+ //
+ // return goodSys;
+ // }
+ //
+ // private static boolean matchSpan(Annotation arg1, Annotation arg2) {
+ // boolean result = false;
+ // result = arg1.getBegin() == arg2.getBegin() && arg1.getEnd() == arg2.getEnd();
+ // return result;
+ // }
+
+ // @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 RemoveEventEventRelations extends JCasAnnotator_ImplBase {
+ 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 relationView;
+ try {
+ relationView = jCas.getView(this.relationViewName);
+ } catch (CASException e) {
+ throw new AnalysisEngineProcessException(e);
+ }
+
+ for(BinaryTextRelation relation : Lists.newArrayList(JCasUtil.select(relationView, BinaryTextRelation.class))){
+ // if(relation.getCategory().equals("CONTAINS")){
+ RelationArgument arg1 = relation.getArg1();
+ RelationArgument arg2 = relation.getArg2();
+ if(arg1.getArgument() instanceof TimeMention && arg2.getArgument() instanceof EventMention ||
+ arg1.getArgument() instanceof EventMention && arg2.getArgument() instanceof TimeMention){
+ // these are the kind we keep.
+ continue;
+ }
+ // if(arg1.getArgument() instanceof EventMention && arg2.getArgument() instanceof EventMention){
+ arg1.removeFromIndexes();
+ arg2.removeFromIndexes();
+ relation.removeFromIndexes();
+ }
+ // }
+
+ }
+ }
+
+ public static class AddPotentialRelations extends JCasAnnotator_ImplBase {
+ 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 relationView;
+ try {
+ relationView = jCas.getView(this.relationViewName);
+ } catch (CASException e) {
+ throw new AnalysisEngineProcessException(e);
+ }
+
+ Map<EventMention, Collection<EventMention>> coveringMap =
+ JCasUtil.indexCovering(relationView, EventMention.class, EventMention.class);
+ for(TemporalTextRelation relation : Lists.newArrayList(JCasUtil.select(relationView, TemporalTextRelation.class))){
+ Annotation arg1 = relation.getArg1().getArgument();
+ Annotation arg2 = relation.getArg2().getArgument();
+ EventMention event = null;
+ if(arg1 instanceof EventMention && arg2 instanceof TimeMention){
+ event = (EventMention) arg1;
+ Collection<EventMention> eventList = coveringMap.get(event);
+ for(EventMention covEvent : eventList){
+ if(!covEvent.getClass().equals(EventMention.class)){
+ createRelation(relationView, covEvent, arg2, relation.getCategory());
+ }
+ }
+ }else if(arg2 instanceof EventMention && arg1 instanceof TimeMention){
+ event = (EventMention) arg2;
+ Collection<EventMention> eventList = coveringMap.get(event);
+ for(EventMention covEvent : eventList){
+ if(!covEvent.getClass().equals(EventMention.class)){
+ createRelation(relationView, arg1, covEvent, relation.getCategory());
+ }
+ }
+ }
+ }
+
+ }
+
+ private 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();
+
+ }
+ }
+
+ /* 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 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 CountSentenceContainsTimes extends 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);
+ }
+
+ //count how many sentences have timex, and how many sentences have only one timex
+ for (Sentence sentence : JCasUtil.select(sentenceView, Sentence.class)) {
+ List<TimeMention> tmentions = JCasUtil.selectCovered(relationView, TimeMention.class, sentence);
+ int timeCount = tmentions==null? 0: tmentions.size();
+ if(timeCount > 0){
+ sentHasTimeCount ++;
+ if(timeCount == 1){
+ sentHasOneTimeCount ++;
+ }
+ }
+ }
+ }
+ }*/
+
+ public static class CountCloseRelation extends JCasAnnotator_ImplBase {
+
+ private String systemViewName = CAS.NAME_DEFAULT_SOFA;
+
+ @Override
+ public void process(JCas jCas) throws AnalysisEngineProcessException {
+ JCas systemView, goldView;
+ int sizeLimit = 6;
+ try {
+ systemView = jCas.getView(this.systemViewName);
+ goldView = jCas.getView(GOLD_VIEW_NAME);
+ } catch (CASException e) {
+ throw new AnalysisEngineProcessException(e);
+ }
+
+ //count how many sentences have timex, and how many sentences have only one timex
+ for (TemporalTextRelation relation : JCasUtil.select(systemView, TemporalTextRelation.class)) {
+ sysRelationCount ++;
+ Annotation arg1 = relation.getArg1().getArgument();
+ Annotation arg2 = relation.getArg2().getArgument();
+ if( arg1.getBegin()> arg2.getBegin()){
+ Annotation temp = arg1;
+ arg1 = arg2;
+ arg2 = temp;
+ }
+ List<WordToken> words = JCasUtil.selectBetween(systemView, WordToken.class, arg1, arg2);
+ if(words.size()<sizeLimit){
+ closeRelationCount++;
+ }
+ }
+
+ Map<List<Annotation>, TemporalTextRelation> relationLookup = new HashMap<>();
+ for (TemporalTextRelation relation : Lists.newArrayList(JCasUtil.select(goldView, TemporalTextRelation.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)){
+ relationLookup.put(key, relation);
+ }
+ }
+
+ //count how many sentences have timex, and how many sentences have only one timex
+ for (TemporalTextRelation relation : relationLookup.values()) {
+ goldRelationCount ++;
+ Annotation arg1 = relation.getArg1().getArgument();
+ Annotation arg2 = relation.getArg2().getArgument();
+ if( arg1.getBegin()> arg2.getBegin()){
+ Annotation temp = arg1;
+ arg1 = arg2;
+ arg2 = temp;
+ }
+ List<WordToken> words = JCasUtil.selectBetween(systemView, WordToken.class, arg1, arg2);
+ if(words.size()<sizeLimit){
+ closeGoldRelationCount++;
+ }
+ }
+ }
+ }
+
+ public static class RemoveRelations extends 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 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 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 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);
+ }
+
+ }
+ }
+ }
+
+ /**
+ * For conflicting temporal relations on the same pair of arguments, if some of the relation types are "overlap", and "contains", remove "overlap" and keep "contains"
+ * @author CH151862
+ *
+ */
+ public static class Overlap2Contains extends JCasAnnotator_ImplBase {
+
+ 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 relationView;
+ try {
+ relationView = jCas.getView(this.relationViewName);
+ } catch (CASException e) {
+ throw new AnalysisEngineProcessException(e);
+ }
+ Map<List<Annotation>, BinaryTextRelation> relationLookup;
+ relationLookup = new HashMap<>();
+ for (BinaryTextRelation relation : Lists.newArrayList(JCasUtil.select(relationView, BinaryTextRelation.class))) {
+ Annotation arg1 = relation.getArg1().getArgument();
+ Annotation arg2 = relation.getArg2().getArgument();
+ String relationType = relation.getCategory();
+ // 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)){
+ BinaryTextRelation storedRel = relationLookup.get(key);
+ String reln = storedRel.getCategory();
+ if(!relationType.equals(reln)){//if there is category conflicts
+ if(relationType.startsWith("OVERLAP") ){//current relation is too general, remove it
+ relation.getArg1().removeFromIndexes();
+ relation.getArg2().removeFromIndexes();
+ relation.removeFromIndexes();
+ }else if( reln.startsWith("OVERLAP") ){//stored relation is too general, remove it
+ //remove duplicate:
+ storedRel.getArg1().removeFromIndexes();
+ storedRel.getArg2().removeFromIndexes();
+ storedRel.removeFromIndexes();
+ relationLookup.put(key, relation);
+ }else if(relationType.startsWith("CONTAINS")){//contain is dominant
+ storedRel.getArg1().removeFromIndexes();
+ storedRel.getArg2().removeFromIndexes();
+ storedRel.removeFromIndexes(jCas);
+ relationLookup.put(key, relation);
+ }else if(reln.startsWith("CONTAINS")){
+ relation.getArg1().removeFromIndexes();
+ relation.getArg2().removeFromIndexes();
+ relation.removeFromIndexes(jCas);
+ }else{
+ relation.getArg1().removeFromIndexes();
+ relation.getArg2().removeFromIndexes();
+ relation.removeFromIndexes();
+ }
+ }else{//if there is no conflicting, remove duplicating relations
+ relation.getArg1().removeFromIndexes();
+ relation.getArg2().removeFromIndexes();
+ relation.removeFromIndexes(jCas);
+ }
+ }else{//if the relation is new, then added it to lookup
+ relationLookup.put(key, relation);
+ }
+
+ }
+
+ }
+ }
+
+ 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);
+ // }
+ // }
+ // }
+ // }
+ //
+ // }
+}
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringFeatureDataWriter.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringFeatureDataWriter.java?rev=1753407&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringFeatureDataWriter.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringFeatureDataWriter.java Tue Jul 19 17:14:35 2016
@@ -0,0 +1,29 @@
+package org.apache.ctakes.temporal.keras;
+
+import java.io.File;
+import java.io.FileNotFoundException;
+
+import com.google.common.annotations.Beta;
+
+/**
+ * <br>
+ * Copyright (c) 2016, Regents of the University of Colorado <br>
+ * All rights reserved.
+ *
+ * @author Tim Miller
+ * @version 2.0.1
+ *
+ */
+@Beta
+public class KerasStringFeatureDataWriter extends ScriptStringOutcomeDataWriter<KerasStringOutcomeClassifierBuilder>{
+
+ public KerasStringFeatureDataWriter(File outputDirectory)
+ throws FileNotFoundException {
+ super(outputDirectory);
+ }
+
+ @Override
+ protected KerasStringOutcomeClassifierBuilder newClassifierBuilder() {
+ return new KerasStringOutcomeClassifierBuilder();
+ }
+}
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeClassifier.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeClassifier.java?rev=1753407&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeClassifier.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeClassifier.java Tue Jul 19 17:14:35 2016
@@ -0,0 +1,29 @@
+package org.apache.ctakes.temporal.keras;
+
+import java.io.File;
+
+import org.cleartk.ml.encoder.features.FeaturesEncoder;
+import org.cleartk.ml.encoder.outcome.OutcomeEncoder;
+import org.cleartk.ml.util.featurevector.FeatureVector;
+
+import com.google.common.annotations.Beta;
+
+/**
+ * <br>
+ * Copyright (c) 2016, Regents of the University of Colorado <br>
+ * All rights reserved.
+ *
+ * @author Tim Miller
+ * @version 2.0.1
+ *
+ */
+@Beta
+public class KerasStringOutcomeClassifier extends ScriptStringOutcomeClassifier {
+
+ public KerasStringOutcomeClassifier(
+ FeaturesEncoder<FeatureVector> featuresEncoder,
+ OutcomeEncoder<String, Integer> outcomeEncoder, File modelDir,
+ File scriptDir) {
+ super(featuresEncoder, outcomeEncoder, modelDir, scriptDir);
+ }
+}
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeClassifierBuilder.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeClassifierBuilder.java?rev=1753407&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeClassifierBuilder.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeClassifierBuilder.java Tue Jul 19 17:14:35 2016
@@ -0,0 +1,87 @@
+package org.apache.ctakes.temporal.keras;
+
+import java.io.File;
+import java.io.IOException;
+import java.util.jar.JarInputStream;
+import java.util.jar.JarOutputStream;
+
+import org.apache.uima.util.Level;
+import org.cleartk.ml.jar.JarStreams;
+
+import com.google.common.annotations.Beta;
+import com.google.common.io.Files;
+
+/**
+ * <br>
+ * Copyright (c) 2016, Regents of the University of Colorado <br>
+ * All rights reserved.
+ *
+ * @author Tim Miller
+ * @version 2.0.1
+ *
+ */
+@Beta
+public class KerasStringOutcomeClassifierBuilder extends ScriptStringOutcomeClassifierBuilder<ScriptStringOutcomeClassifier> {
+
+ @Override
+ public void packageClassifier(File dir, JarOutputStream modelStream) throws IOException {
+ super.packageClassifier(dir, modelStream);
+
+ JarStreams.putNextJarEntry(modelStream, "outcome-lookup.txt", new File(dir, "outcome-lookup.txt"));
+
+ int modelNum = 0;
+ while(true){
+ File modelArchFile = new File(dir, getArchFilename(modelNum));
+ File modelWeightsFile = new File(dir, getWeightsFilename(modelNum));
+ if(!modelArchFile.exists()) break;
+
+ JarStreams.putNextJarEntry(modelStream, modelArchFile.getName(), modelArchFile.getAbsoluteFile());
+ JarStreams.putNextJarEntry(modelStream, modelWeightsFile.getName(), modelWeightsFile.getAbsoluteFile());
+ modelNum++;
+ }
+ }
+
+ @Override
+ protected void unpackageClassifier(JarInputStream modelStream)
+ throws IOException {
+ super.unpackageClassifier(modelStream);
+
+
+ // create the model dir to unpack all the model files
+ this.modelDir = Files.createTempDir();
+
+ // grab the script dir from the manifest:
+ this.scriptDir = new File(modelStream.getManifest().getMainAttributes().getValue(SCRIPT_DIR_PARAM));
+
+ extractFileToDir(modelDir, modelStream, "outcome-lookup.txt");
+
+ int modelNum = 0;
+ while(true){
+ String archFn = getArchFilename(modelNum);
+ String wtsFn = getWeightsFilename(modelNum);
+
+ try{
+ extractFileToDir(modelDir, modelStream, archFn);
+ extractFileToDir(modelDir, modelStream, wtsFn);
+ }catch(IOException e){
+ logger.log(Level.WARNING, "Encountered the following exception: " + e.getMessage());
+ break;
+ }
+ modelNum++;
+ }
+ }
+
+
+ @Override
+ protected KerasStringOutcomeClassifier newClassifier() {
+ return new KerasStringOutcomeClassifier(this.featuresEncoder, this.outcomeEncoder, this.modelDir, this.scriptDir);
+ }
+
+ private static String getArchFilename(int num){
+ return "model_"+num+".json";
+ }
+
+ private static String getWeightsFilename(int num){
+ return "model_"+num+".h5";
+ }
+}
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeDataWriter.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeDataWriter.java?rev=1753407&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeDataWriter.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/KerasStringOutcomeDataWriter.java Tue Jul 19 17:14:35 2016
@@ -0,0 +1,38 @@
+package org.apache.ctakes.temporal.keras;
+
+import java.io.File;
+import java.io.FileNotFoundException;
+
+import org.cleartk.ml.CleartkProcessingException;
+import org.cleartk.ml.util.featurevector.FeatureVector;
+
+import com.google.common.annotations.Beta;
+
+/**
+ * <br>
+ * Copyright (c) 2016, Regents of the University of Colorado <br>
+ * All rights reserved.
+ *
+ * @author Tim Miller
+ * @version 2.0.1
+ *
+ */
+@Beta
+public class KerasStringOutcomeDataWriter extends ScriptStringFeatureDataWriter<KerasStringOutcomeClassifierBuilder>{
+
+ public KerasStringOutcomeDataWriter(File outputDirectory)
+ throws FileNotFoundException {
+ super(outputDirectory);
+ }
+
+ @Override
+ protected KerasStringOutcomeClassifierBuilder newClassifierBuilder() {
+ return new KerasStringOutcomeClassifierBuilder();
+ }
+
+ @Override
+ protected void writeEncoded(FeatureVector features, Integer outcome) throws CleartkProcessingException {
+ // TODO Auto-generated method stub
+
+ }
+}
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringFeatureDataWriter.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringFeatureDataWriter.java?rev=1753407&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringFeatureDataWriter.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringFeatureDataWriter.java Tue Jul 19 17:14:35 2016
@@ -0,0 +1,65 @@
+package org.apache.ctakes.temporal.keras;
+
+import java.io.File;
+import java.io.FileNotFoundException;
+//import java.util.Locale;
+import java.util.List;
+
+import org.apache.uima.UimaContext;
+import org.apache.uima.fit.descriptor.ConfigurationParameter;
+import org.apache.uima.fit.factory.initializable.Initializable;
+import org.apache.uima.resource.ResourceInitializationException;
+import org.cleartk.ml.CleartkProcessingException;
+import org.cleartk.ml.Feature;
+import org.cleartk.ml.Instance;
+import org.cleartk.ml.encoder.features.BooleanEncoder;
+import org.cleartk.ml.encoder.features.FeatureVectorFeaturesEncoder;
+import org.cleartk.ml.encoder.features.NumberEncoder;
+import org.cleartk.ml.encoder.features.StringEncoder;
+import org.cleartk.ml.encoder.outcome.StringToIntegerOutcomeEncoder;
+import org.cleartk.ml.jar.DataWriter_ImplBase;
+import org.cleartk.ml.util.featurevector.FeatureVector;
+
+public abstract class ScriptStringFeatureDataWriter<T extends ScriptStringOutcomeClassifierBuilder<ScriptStringOutcomeClassifier>>
+ extends DataWriter_ImplBase<T, FeatureVector, String,Integer> implements Initializable {
+
+ public static final String PARAM_SCRIPT_DIR = "DataWriterScriptDirectory";
+ @ConfigurationParameter(name=PARAM_SCRIPT_DIR)
+ public String dir;
+
+ public ScriptStringFeatureDataWriter(File outputDirectory)
+ throws FileNotFoundException {
+ super(outputDirectory);
+ FeatureVectorFeaturesEncoder fe = new FeatureVectorFeaturesEncoder();
+ fe.addEncoder(new NumberEncoder());
+ fe.addEncoder(new BooleanEncoder());
+ fe.addEncoder(new StringEncoder());
+ this.setFeaturesEncoder(fe);
+ this.setOutcomeEncoder(new StringToIntegerOutcomeEncoder());
+ }
+
+ @Override
+ public void write(Instance<String> instance)
+ throws CleartkProcessingException {
+ this.trainingDataWriter.print(instance.getOutcome());
+ this.trainingDataWriter.print("|");
+// for (Feature feat : instance.getFeatures()) {
+// this.trainingDataWriter.print(" " + feat.getValue());
+// }
+ List<Feature> features = instance.getFeatures();
+ for (int i = 0; i < features.size(); i ++){
+ this.trainingDataWriter.print(features.get(i).getValue());
+ if( i < features.size()-1){
+ this.trainingDataWriter.print(" ");
+ }
+ }
+ this.trainingDataWriter.println();
+ }
+
+ @Override
+ public void initialize(UimaContext context)
+ throws ResourceInitializationException {
+ this.dir = (String) context.getConfigParameterValue(PARAM_SCRIPT_DIR);
+ this.classifierBuilder.setScriptDirectory(this.dir);
+ }
+}
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeClassifier.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeClassifier.java?rev=1753407&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeClassifier.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeClassifier.java Tue Jul 19 17:14:35 2016
@@ -0,0 +1,123 @@
+package org.apache.ctakes.temporal.keras;
+
+import java.io.BufferedReader;
+import java.io.File;
+import java.io.IOException;
+import java.io.InputStreamReader;
+import java.io.PrintStream;
+import java.util.List;
+import java.util.Locale;
+
+import org.apache.uima.UIMAFramework;
+import org.apache.uima.util.Level;
+import org.apache.uima.util.Logger;
+import org.cleartk.ml.CleartkProcessingException;
+import org.cleartk.ml.Feature;
+import org.cleartk.ml.encoder.features.FeaturesEncoder;
+import org.cleartk.ml.encoder.outcome.OutcomeEncoder;
+import org.cleartk.ml.jar.Classifier_ImplBase;
+import org.cleartk.ml.util.featurevector.FeatureVector;
+
+import com.google.common.annotations.Beta;
+
+/**
+ * <br>
+ * Copyright (c) 2016, Regents of the University of Colorado <br>
+ * All rights reserved.
+ *
+ * @author Tim Miller
+ * @version 2.0.1
+ *
+ */
+@Beta
+public abstract class ScriptStringOutcomeClassifier extends Classifier_ImplBase<FeatureVector, String, Integer> {
+ File modelDir = null;
+ Process classifierProcess = null;
+ PrintStream toClassifier = null;
+ BufferedReader reader = null;
+ BufferedReader errReader = null;
+ Logger logger = UIMAFramework.getLogger(ScriptStringOutcomeClassifier.class);
+
+ public ScriptStringOutcomeClassifier(
+ FeaturesEncoder<FeatureVector> featuresEncoder,
+ OutcomeEncoder<String, Integer> outcomeEncoder,
+ File modelDir,
+ File scriptDir) {
+ super(featuresEncoder, outcomeEncoder);
+ this.modelDir = modelDir;
+
+ File classifyScript = null;
+ for(File file : scriptDir.listFiles()){
+ if(file.getName().startsWith("classify.sh")){
+ if(classifyScript != null){
+ throw new RuntimeException("There are multiple files named classify.*");
+ }
+ classifyScript = file;
+ }
+ }
+
+ if(classifyScript == null){
+ throw new RuntimeException("There are no files named classify.*");
+ }
+
+ try {
+ this.classifierProcess = Runtime.getRuntime().exec(new String[]{
+ classifyScript.getAbsolutePath(),
+ modelDir.getAbsolutePath()});
+ // start the classifier process running, give it a chance to read the model, and
+ // set classifierProcess to the running classifier
+ toClassifier = new PrintStream(classifierProcess.getOutputStream());
+ reader = new BufferedReader(new InputStreamReader(classifierProcess.getInputStream()));
+ errReader = new BufferedReader(new InputStreamReader(classifierProcess.getErrorStream()));
+ errReader.readLine(); // read line about which backend it is using.
+ } catch (IOException e) {
+ e.printStackTrace();
+ throw new RuntimeException(e);
+ }
+ }
+
+ public String classify(List<Feature> features)
+ throws CleartkProcessingException {
+ // Encode the features and pass them to the standard input of the classifier process
+ // and then read the standard output prediction, which will be in the string format expected by
+ // the annotator.
+
+ StringBuilder buf = new StringBuilder();
+
+// for (FeatureVector.Entry featureNode : this.featuresEncoder.encodeAll(features)) {
+// buf.append(String.format(Locale.US, " %d:%.7f", featureNode.index, featureNode.value));
+// }
+ for (int i = 0; i < features.size(); i ++){
+ buf.append(features.get(i).getValue());
+ if( i < features.size()-1){
+ buf.append(" ");
+ }
+ }
+
+ this.toClassifier.println(buf);
+ this.toClassifier.flush();
+
+ String line = "";
+ String eLine = "";
+ try {
+ line = reader.readLine();
+ if(line == null){
+ while((eLine = errReader.readLine()) != null){
+ logger.log(Level.SEVERE, eLine);
+ }
+ }
+ } catch (IOException e) {
+ throw new RuntimeException(e);
+ }
+
+ return line;
+ }
+
+ @Override
+ protected void finalize() throws Throwable {
+ super.finalize();
+
+ this.toClassifier.print('\n');
+ classifierProcess.waitFor();
+ }
+}
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeClassifierBuilder.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeClassifierBuilder.java?rev=1753407&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeClassifierBuilder.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeClassifierBuilder.java Tue Jul 19 17:14:35 2016
@@ -0,0 +1,103 @@
+package org.apache.ctakes.temporal.keras;
+
+import java.io.BufferedReader;
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.io.InputStreamReader;
+import java.util.jar.Attributes;
+import java.util.jar.JarInputStream;
+
+import org.apache.uima.UIMAFramework;
+import org.apache.uima.util.Level;
+import org.apache.uima.util.Logger;
+import org.cleartk.ml.jar.ClassifierBuilder_ImplBase;
+import org.cleartk.ml.jar.JarStreams;
+import org.cleartk.ml.util.featurevector.FeatureVector;
+
+
+public abstract class ScriptStringOutcomeClassifierBuilder<T extends ScriptStringOutcomeClassifier> extends ClassifierBuilder_ImplBase<T, FeatureVector, String, Integer> {
+ public static final Attributes.Name SCRIPT_DIR_PARAM = new Attributes.Name("ScriptDirectory");
+ protected static final Logger logger = UIMAFramework.getLogger(ScriptStringOutcomeClassifierBuilder.class);
+
+ protected File modelDir = null;
+ protected File scriptDir = null;
+
+ public void setScriptDirectory(String scriptDir){
+ Attributes atts = this.manifest.getMainAttributes();
+ atts.put(SCRIPT_DIR_PARAM, scriptDir);
+ }
+
+ @Override
+ public File getTrainingDataFile(File dir) {
+ return new File(dir, "training-data.liblinear");
+ }
+
+ @Override
+ public void trainClassifier(File dir, String... args) throws Exception {
+ // args[0] should be path to directory with model training code:
+ // args[1-] is the set of arguments the program takes.
+ // dir is by convention the first argument that the training script takes.
+
+ if(this.scriptDir == null){
+ this.scriptDir = new File(this.manifest.getMainAttributes().getValue(SCRIPT_DIR_PARAM));
+ }
+ // first find the train script:
+ File trainScript = null;
+
+ for(File file : this.scriptDir.listFiles()){
+ if(file.getName().startsWith("train.")){//calls train.sh
+ if(trainScript != null){
+ throw new RuntimeException("There are multiple files named train.*");
+ }
+ trainScript = file;
+ }
+ }
+ if(trainScript == null) throw new RuntimeException("ERROR: Train directory does not contain any scripts named train.*");
+ StringBuilder cmdArgs = new StringBuilder();
+ for(int i = 0; i < args.length; i++){
+ cmdArgs.append(args[i]);
+ cmdArgs.append(' ');
+ }
+ String arg2 = "";
+ if(cmdArgs.length() > 0){
+ arg2 = cmdArgs.substring(0, cmdArgs.length()-1);
+ }
+ Process p = Runtime.getRuntime().exec(new String[]{
+ trainScript.getAbsolutePath(),
+ dir.getAbsolutePath(),
+ arg2
+ });
+
+ String line = "";
+
+ BufferedReader reader =
+ new BufferedReader(new InputStreamReader(p.getInputStream()));
+ while ((line = reader.readLine()) != null) {
+ logger.log(Level.INFO, line);
+ }
+ BufferedReader errReader = new BufferedReader(new InputStreamReader(p.getErrorStream()));
+ while((line = errReader.readLine()) != null){
+ logger.log(Level.WARNING, line);
+ }
+ int ret = p.waitFor();
+ if(ret != 0){
+ throw new RuntimeException();
+ }
+ }
+
+ protected static void extractFileToDir(File dir, JarInputStream modelStream, String fn) throws IOException{
+ JarStreams.getNextJarEntry(modelStream, fn);
+ File outFile = new File(dir, fn);
+ try(FileOutputStream fos = new FileOutputStream(outFile)){
+ byte[] byteArray = new byte[1024];
+ int i;
+ while ((i = modelStream.read(byteArray)) > 0)
+ {
+ //Write the bytes to the output stream
+ fos.write(byteArray, 0, i);
+ }
+ }
+ }
+
+}
Added: ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeDataWriter.java
URL: http://svn.apache.org/viewvc/ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeDataWriter.java?rev=1753407&view=auto
==============================================================================
--- ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeDataWriter.java (added)
+++ ctakes/trunk/ctakes-temporal/src/main/java/org/apache/ctakes/temporal/keras/ScriptStringOutcomeDataWriter.java Tue Jul 19 17:14:35 2016
@@ -0,0 +1,54 @@
+package org.apache.ctakes.temporal.keras;
+
+import java.io.File;
+import java.io.FileNotFoundException;
+import java.util.Locale;
+
+import org.apache.uima.UimaContext;
+import org.apache.uima.fit.descriptor.ConfigurationParameter;
+import org.apache.uima.fit.factory.initializable.Initializable;
+import org.apache.uima.resource.ResourceInitializationException;
+import org.cleartk.ml.CleartkProcessingException;
+import org.cleartk.ml.encoder.features.BooleanEncoder;
+import org.cleartk.ml.encoder.features.FeatureVectorFeaturesEncoder;
+import org.cleartk.ml.encoder.features.NumberEncoder;
+import org.cleartk.ml.encoder.features.StringEncoder;
+import org.cleartk.ml.encoder.outcome.StringToIntegerOutcomeEncoder;
+import org.cleartk.ml.jar.DataWriter_ImplBase;
+import org.cleartk.ml.util.featurevector.FeatureVector;
+
+public abstract class ScriptStringOutcomeDataWriter<T extends ScriptStringOutcomeClassifierBuilder<ScriptStringOutcomeClassifier>>
+ extends DataWriter_ImplBase<T, FeatureVector, String,Integer> implements Initializable {
+
+ public static final String PARAM_SCRIPT_DIR = "DataWriterScriptDirectory";
+ @ConfigurationParameter(name=PARAM_SCRIPT_DIR)
+ public String dir;
+
+ public ScriptStringOutcomeDataWriter(File outputDirectory)
+ throws FileNotFoundException {
+ super(outputDirectory);
+ FeatureVectorFeaturesEncoder fe = new FeatureVectorFeaturesEncoder();
+ fe.addEncoder(new NumberEncoder());
+ fe.addEncoder(new BooleanEncoder());
+ fe.addEncoder(new StringEncoder());
+ this.setFeaturesEncoder(fe);
+ this.setOutcomeEncoder(new StringToIntegerOutcomeEncoder());
+ }
+
+ @Override
+ protected void writeEncoded(FeatureVector features, Integer outcome)
+ throws CleartkProcessingException {
+ this.trainingDataWriter.print(outcome);
+ for (FeatureVector.Entry featureNode : features) {
+ this.trainingDataWriter.format(Locale.US, " %d:%.7f", featureNode.index, featureNode.value);
+ }
+ this.trainingDataWriter.println();
+ }
+
+ @Override
+ public void initialize(UimaContext context)
+ throws ResourceInitializationException {
+ this.dir = (String) context.getConfigParameterValue(PARAM_SCRIPT_DIR);
+ this.classifierBuilder.setScriptDirectory(this.dir);
+ }
+}