Author: schor
Date: Tue Sep 14 15:34:14 2010
New Revision: 996953
URL: http://svn.apache.org/viewvc?rev=996953&view=rev
Log:
[UIMA-1833] committed the update, which adds an xml descriptor and an annotator
that extracts tokens from a CAS to use as input to the existing HMM training
code.
Added:
uima/sandbox/trunk/Tagger/desc/HmmModelTrainer.xml
uima/sandbox/trunk/Tagger/src/main/java/org/apache/uima/examples/tagger/HMMModelTrainer.java
Modified:
uima/sandbox/trunk/Tagger/src/main/java/org/apache/uima/examples/tagger/trainAndTest/ModelGeneration.java
Added: uima/sandbox/trunk/Tagger/desc/HmmModelTrainer.xml
URL:
http://svn.apache.org/viewvc/uima/sandbox/trunk/Tagger/desc/HmmModelTrainer.xml?rev=996953&view=auto
==============================================================================
--- uima/sandbox/trunk/Tagger/desc/HmmModelTrainer.xml (added)
+++ uima/sandbox/trunk/Tagger/desc/HmmModelTrainer.xml Tue Sep 14 15:34:14 2010
@@ -0,0 +1,82 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<analysisEngineDescription xmlns="http://uima.apache.org/resourceSpecifier";>
+ <frameworkImplementation>org.apache.uima.java</frameworkImplementation>
+ <primitive>true</primitive>
+
<annotatorImplementationName>org.apache.uima.examples.tagger.HMMModelTrainer</annotatorImplementationName>
+ <analysisEngineMetaData>
+ <name>HMMModelTrainer</name>
+ <description>This analysis engine trains an N-gram model for the HMM
tagger. It uses a training corpus as reference. This corpus must contain
annotations on words with an attribute corresponding of the POS value to be
learned.
+
+The configuration of this analysis engine is done through several parameters:
+<ul>
+<li>View: - the view from which the tokens will be extracted</li>
+<li>ModelExportFile: - the path where the model will be
written</li>
+<li>FeaturePathPOS: - feature path to the value of the POS to be
learned. The annotation should exactly cover a "word".</li>
+</ul>
+
+<b>BEWARE: this analysis engine does not allow multiple deployment
!</b>
+
+<i>NB. At the moment: both bi and trigram statistics are saved in one
model file.</i></description>
+ <version>1.0</version>
+ <vendor/>
+ <configurationParameters>
+ <configurationParameter>
+ <name>View</name>
+ <description>The view from which the tokens will be
extracted.</description>
+ <type>String</type>
+ <multiValued>false</multiValued>
+ <mandatory>true</mandatory>
+ </configurationParameter>
+ <configurationParameter>
+ <name>ModelExportFile</name>
+ <description>The path where the model will be written.</description>
+ <type>String</type>
+ <multiValued>false</multiValued>
+ <mandatory>true</mandatory>
+ </configurationParameter>
+ <configurationParameter>
+ <name>FeaturePathPOS</name>
+ <description>Feature path to the value of the POS to be learnt. The
annotation should exactly cover a "word".</description>
+ <type>String</type>
+ <multiValued>false</multiValued>
+ <mandatory>true</mandatory>
+ </configurationParameter>
+ </configurationParameters>
+ <configurationParameterSettings>
+ <nameValuePair>
+ <name>View</name>
+ <value>
+ <string>_InitialView</string>
+ </value>
+ </nameValuePair>
+ <nameValuePair>
+ <name>ModelExportFile</name>
+ <value>
+ <string>hmmtagger_model.dat</string>
+ </value>
+ </nameValuePair>
+ <nameValuePair>
+ <name>FeaturePathPOS</name>
+ <value>
+ <string>org.apache.uima.TokenAnnotation:posTag</string>
+ </value>
+ </nameValuePair>
+ </configurationParameterSettings>
+ <typeSystemDescription/>
+ <typePriorities/>
+ <fsIndexCollection/>
+ <capabilities>
+ <capability>
+ <inputs/>
+ <outputs/>
+ <languagesSupported/>
+ </capability>
+ </capabilities>
+ <operationalProperties>
+ <modifiesCas>false</modifiesCas>
+ <multipleDeploymentAllowed>false</multipleDeploymentAllowed>
+ <outputsNewCASes>false</outputsNewCASes>
+ </operationalProperties>
+ </analysisEngineMetaData>
+ <resourceManagerConfiguration/>
+</analysisEngineDescription>
Added:
uima/sandbox/trunk/Tagger/src/main/java/org/apache/uima/examples/tagger/HMMModelTrainer.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/trunk/Tagger/src/main/java/org/apache/uima/examples/tagger/HMMModelTrainer.java?rev=996953&view=auto
==============================================================================
---
uima/sandbox/trunk/Tagger/src/main/java/org/apache/uima/examples/tagger/HMMModelTrainer.java
(added)
+++
uima/sandbox/trunk/Tagger/src/main/java/org/apache/uima/examples/tagger/HMMModelTrainer.java
Tue Sep 14 15:34:14 2010
@@ -0,0 +1,186 @@
+/*
+ * 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.uima.examples.tagger;
+
+
+// Java dependencies
+import java.util.ArrayList;
+// UIMA dependencies
+import org.apache.uima.UIMAFramework;
+import org.apache.uima.UimaContext;
+import org.apache.uima.analysis_component.JCasAnnotator_ImplBase;
+import org.apache.uima.analysis_engine.AnalysisEngineProcessException;
+import org.apache.uima.cas.CASException;
+import org.apache.uima.cas.FSIterator;
+import org.apache.uima.cas.Feature;
+import org.apache.uima.cas.Type;
+import org.apache.uima.examples.tagger.trainAndTest.ModelGeneration;
+import org.apache.uima.examples.tagger.trainAndTest.Token;
+import org.apache.uima.jcas.JCas;
+import org.apache.uima.jcas.tcas.Annotation;
+import org.apache.uima.resource.ResourceInitializationException;
+import org.apache.uima.util.Level;
+
+/**
+ * This analysis engine trains an N-gram model for the HMM tagger. It uses
+ * a training corpus as reference. This corpus must contain annotations on
+ * words with an attribute corresponding to the POS value to be learned.
+ *
+ * The configuration of this analysis engine is done through several
parameters:
+ * <ul>
+ * <li>View: - the view from which the tokens will be extracted</li>
+ * <li>ModelExportFile: - the path where the model will be written</li>
+ * <li>FeaturePathPOS: - feature path to the value of the POS to be learned.
+ * The annotation should exactly cover a "word".</li>
+ * </ul>
+ *
+ * <b>BEWARE: this analysis engine does not allow multiple deployment !</b>
+ *
+ * NB. At the moment: both bi and trigram statistics are saved in one model
file.
+ */
+public class HMMModelTrainer extends JCasAnnotator_ImplBase {
+
+ /** Name of the parameter for the view */
+ public static String PARAM_VIEW = "View";
+ /** Name of the parameter for the model export path */
+ public static String PARAM_FILE = "ModelExportFile";
+ /** Name of the parameter for the feature path to the POS */
+ public static String PARAM_POSFP = "FeaturePathPOS";
+
+ /** The view from which the tokens will be extracted */
+ private String theView;
+ /** The path to the file where the model will be written */
+ private String fileOutput;
+ /** The type from which we will extract the tags to learn */
+ private String theTokenTypeName;
+ /** The name of the attribute where the POS to learn is stored */
+ private String thePOSAttribute;
+
+ /** The list of collected tokens */
+ private ArrayList<Token> theLearnedTokens;
+
+ /**
+ * Initialization of the component
+ */
+ public void initialize(UimaContext aContext)
+ throws ResourceInitializationException {
+ super.initialize(aContext);
+ // Configure the component
+ theView =
+ (String) aContext.getConfigParameterValue(PARAM_VIEW);
+ fileOutput =
+ (String)
aContext.getConfigParameterValue(PARAM_FILE);
+ // Compute the type and the attribute name
+ String fpPOS =
+ (String) aContext.getConfigParameterValue(PARAM_POSFP);
+ Integer idx = fpPOS.lastIndexOf(":");
+ if (idx >= 0) {
+ theTokenTypeName = fpPOS.substring(0, idx);
+ thePOSAttribute = fpPOS.substring(idx+1);
+ } else {
+ throw new ResourceInitializationException("The feature
path passed " +
+ "in parameter ('"+fpPOS+"') is not
valid. " +
+ "It should be like :
'type.name:attribute'", null);
+ }
+ // Prepare the list of tokens
+ theLearnedTokens = new ArrayList<Token>();
+ }
+
+ /**
+ * Processing.
+ * Browse the annotations of the type theTokenTypeName that must
inherit
+ * from the type tcas.Annotation and build the list of tokens that will
be
+ * learned by the HMMTagger.
+ */
+ @Override
+ public void process(JCas cas) throws AnalysisEngineProcessException {
+ try {
+ // Select the view we will work on
+ JCas workingView = cas.getView(theView);
+ // Iterate over the type we will learn from
+ Type tokenType =
+
workingView.getTypeSystem().getType(theTokenTypeName);
+ if (tokenType != null) {
+ // Compute the feature for the POS value
+ Feature featPOS =
+
tokenType.getFeatureByBaseName(thePOSAttribute);
+ // Browse these annotation and create the
tokens to be learned
+ FSIterator<Annotation> itPOS =
+
workingView.getAnnotationIndex(tokenType).iterator();
+ Integer c = 0;
+ Integer i = 0;
+ while ( itPOS.hasNext() ) {
+ Annotation token = itPOS.next();
+ // Create a new token to be learned and
add it to the list,
+ // if the POS value is relevant (not
null)
+ Token tokenTmp = new Token();
+ tokenTmp.word = token.getCoveredText();
+ tokenTmp.pos =
token.getStringValue(featPOS);
+ if (tokenTmp.pos != null) {
+ c++;
+ theLearnedTokens.add(tokenTmp);
+ } else {
+
UIMAFramework.getLogger().log(Level.WARNING,
+ "Ignoring token
"+tokenTmp.word+" because its " +
+ "POS value is
null");
+ i++;
+ }
+ }
+ // Log the number of tokens collected
+ UIMAFramework.getLogger().log(Level.INFO,
+ c + " tokens to be learned
added, " + i +
+ " tokens ignored.");
+ } else {
+ throw new AnalysisEngineProcessException(
+ "The type '"+theTokenTypeName+"',
passed as token type " +
+ "for the training is not in the type
system.", null);
+ }
+ } catch (CASException e) {
+ throw new AnalysisEngineProcessException(e);
+ }
+ }
+
+ /**
+ * Called at the end of the processing.
+ * When the whole collection has been processed, we create the model
+ * from the elements we collected.
+ */
+ @Override
+ public void collectionProcessComplete() throws
AnalysisEngineProcessException {
+ try {
+ UIMAFramework.getLogger().log(Level.INFO,
+ "Generation of model '"+fileOutput+"'
with " +
+ theLearnedTokens.size()+" tokens to be
learned.");
+ ModelGeneration md =
+ new
ModelGeneration(theLearnedTokens,fileOutput);
+ md.init();
+ UIMAFramework.getLogger().log(Level.INFO, "Model
generated: " +
+ md.suffix_tree.size() + " leaves suffix tree, " +
+ md.transition_probs.size() + " transitions
probabilities, " +
+ md.word_probs.size() + " tag probabilities.");
+ } catch (Exception e) {
+ // Because UIMA filter all exceptions... we try to
catch those
+ UIMAFramework.getLogger().log(Level.SEVERE,
+ "Something happened : " +
e.getMessage());
+ e.printStackTrace();
+ throw new AnalysisEngineProcessException(e);
+ }
+ }
+}
+
Modified:
uima/sandbox/trunk/Tagger/src/main/java/org/apache/uima/examples/tagger/trainAndTest/ModelGeneration.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/trunk/Tagger/src/main/java/org/apache/uima/examples/tagger/trainAndTest/ModelGeneration.java?rev=996953&r1=996952&r2=996953&view=diff
==============================================================================
---
uima/sandbox/trunk/Tagger/src/main/java/org/apache/uima/examples/tagger/trainAndTest/ModelGeneration.java
(original)
+++
uima/sandbox/trunk/Tagger/src/main/java/org/apache/uima/examples/tagger/trainAndTest/ModelGeneration.java
Tue Sep 14 15:34:14 2010
@@ -90,7 +90,7 @@ public class ModelGeneration implements
}
- private void init(){
+ public void init(){
List<Map<String, Map<String,Double>>> l =
get_word_probs(get_lexicon(corpus));
