/**
* Copyright (c) 2009/09-2012/08, Regents of the University of Colorado
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/**
* Copyright 2012/09-2013/04, University of Massachusetts Amherst
* Copyright 2013/05-Present, IPSoft Inc.
*
* Licensed 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 com.clearnlp.bin;
import java.io.BufferedInputStream;
import java.io.BufferedOutputStream;
import java.io.FileInputStream;
import java.io.FileOutputStream;
import java.io.ObjectInputStream;
import java.io.ObjectOutputStream;
import java.io.PrintStream;
import java.util.ArrayList;
import java.util.List;
import java.util.Set;
import java.util.zip.GZIPInputStream;
import java.util.zip.GZIPOutputStream;
import org.kohsuke.args4j.Option;
import org.w3c.dom.Element;
import org.w3c.dom.NodeList;
import com.clearnlp.classification.algorithm.AbstractAlgorithm;
import com.clearnlp.classification.feature.JointFtrXml;
import com.clearnlp.classification.model.StringModelAD;
import com.clearnlp.collection.list.FloatArrayList;
import com.clearnlp.component.evaluation.AbstractEval;
import com.clearnlp.component.online.AbstractOnlineStatisticalComponent;
import com.clearnlp.component.online.OnlinePOSTagger;
import com.clearnlp.component.state.AbstractState;
import com.clearnlp.dependency.DEPTree;
import com.clearnlp.morphology.MPLib;
import com.clearnlp.nlp.NLPMode;
import com.clearnlp.reader.JointReader;
import com.clearnlp.util.UTArgs4j;
import com.clearnlp.util.UTFile;
import com.clearnlp.util.UTInput;
import com.clearnlp.util.UTOutput;
import com.clearnlp.util.UTXml;
import com.clearnlp.util.map.Prob1DMap;
import com.clearnlp.util.pair.ObjectDoublePair;
import com.google.common.collect.Sets;
/**
* @since 2.0.1
* @author Jinho D. Choi ({@code jdchoi77@gmail.com})
*/
public class NLPDevelop extends AbstractNLP implements NLPMode
{
protected final String DELIM_FILENAME = ":";
protected final int MAX_TREES = 5000;
@Option(name="-c", usage="configuration file (required)", required=true, metaVar="<filename>")
protected String s_configFile;
@Option(name="-f", usage="feature template files delimited by '"+DELIM_FILENAME+"' (required)", required=true, metaVar="<filename>")
protected String s_featureFiles;
@Option(name="-i", usage="training file or directory containing training files (required)", required=true, metaVar="<directory>")
protected String s_trainPath;
@Option(name="-z", usage="mode (pos|dep|pred|role|srl)", required=true, metaVar="<string>")
protected String s_mode;
@Option(name="-d", usage="development file or directory containing development files (required)", required=true, metaVar="<directory>")
protected String s_developPath;
@Option(name="-m", usage="model file (required)", required=true, metaVar="<filename>")
protected String s_modelFile;
@Option(name="-t", usage="type (required)", required=true, metaVar="<0|1|2|3>")
protected int i_type;
// @Option(name="-r", usage="random seed", required=false, metaVar="<integer>")
// protected int i_randomSeed = 11;
public NLPDevelop(String[] args)
{
UTArgs4j.initArgs(this, args);
String[] featureFiles = s_featureFiles.split(DELIM_FILENAME);
String[] trainFiles = UTFile.getSortedFileListBySize(s_trainPath, ".*", true);
String[] developFiles = UTFile.getSortedFileListBySize(s_developPath, ".*", true);
try
{
Element eConfig = UTXml.getDocumentElement(new FileInputStream(s_configFile));
JointFtrXml[] xmls = getFeatureTemplates(featureFiles);
switch (i_type)
{
case 0: develop(xmls, trainFiles, developFiles, eConfig, s_mode, -1); break;
case 1: train(xmls, trainFiles, s_modelFile, eConfig, s_mode); break;
case 2: decode(developFiles, s_modelFile, eConfig);
}
}
catch (Exception e) {e.printStackTrace();}
}
public void decode(String[] inputFiles, String modelFile, Element eConfig) throws Exception
{
JointReader reader = getJointReader(UTXml.getFirstElementByTagName(eConfig, TAG_READER));
AbstractOnlineStatisticalComponent<? extends AbstractState> component = getDecoder(new ObjectInputStream(new BufferedInputStream(new GZIPInputStream(new FileInputStream(modelFile)))));
process(inputFiles, reader, component, "Decoding:", FLAG_DECODE, -1);
}
public void train(JointFtrXml[] xmls, String[] trainFiles, String modelFile, Element eConfig, String mode) throws Exception
{
JointReader reader = getJointReader(UTXml.getFirstElementByTagName(eConfig, TAG_READER));
AbstractOnlineStatisticalComponent<? extends AbstractState> component = preBootstrap(xmls, trainFiles, reader, eConfig, mode, -1);
Element eMode = UTXml.getFirstElementByTagName(eConfig, mode);
NodeList eTrains = eMode.getElementsByTagName(TAG_TRAIN);
int boot = 0, nBootstraps = getNumberOfBootstraps(eMode);
while (true)
{
train(component, eTrains, boot);
if (boot >= nBootstraps) break;
LOG.info(String.format("===== Bootstrap: %d =====\n", ++boot));
process(trainFiles, reader, component, "Generating instances:", FLAG_BOOTSTRAP, -1);
}
for (StringModelAD model : component.getModels())
model.trimFeatures(LOG, 0f);
component.save(new ObjectOutputStream(new BufferedOutputStream(new GZIPOutputStream(new FileOutputStream(modelFile)))));
}
public void develop(JointFtrXml[] xmls, String[] trainFiles, String[] developFiles, Element eConfig, String mode, int devId) throws Exception
{
JointReader reader = getJointReader(UTXml.getFirstElementByTagName(eConfig, TAG_READER));
AbstractOnlineStatisticalComponent<? extends AbstractState> component = preBootstrap(xmls, trainFiles, reader, eConfig, mode, devId);
Element eMode = UTXml.getFirstElementByTagName(eConfig, mode);
NodeList eTrains = eMode.getElementsByTagName(TAG_TRAIN);
double currScore, bestScore = 0;
int boot = 0;
while (true)
{
if (boot == 0)
{
develop(trainFiles, reader, component, eTrains, getBootstrapScore(eMode), boot, FLAG_EVALUATE);
}
else
{
currScore = develop(developFiles, reader, component, eTrains, 0d, boot, FLAG_EVALUATE);
if (currScore <= bestScore) break;
bestScore = currScore;
}
LOG.info(String.format("===== Bootstrap: %d =====\n", ++boot));
process(trainFiles, reader, component, "Generating instances:", FLAG_BOOTSTRAP, devId);
}
}
private AbstractOnlineStatisticalComponent<? extends AbstractState> preBootstrap(JointFtrXml[] xmls, String[] trainFiles, JointReader reader, Element eConfig, String mode, int devId) throws Exception
{
AbstractOnlineStatisticalComponent<? extends AbstractState> component;
Object[] lexica = null;
// collect
component = getCollector(xmls, trainFiles, reader, eConfig, devId);
if (component != null)
{
process(trainFiles, reader, component, "Collecting lexica:", FLAG_COLLECT, devId);
lexica = component.getLexica();
}
// train
component = getTrainer(xmls, lexica);
process(trainFiles, reader, component, "Generating instances:", FLAG_TRAIN, devId);
return component;
}
// ================================== PROCESS ==================================
protected List<String> process(String[] filenames, JointReader reader, AbstractOnlineStatisticalComponent<? extends AbstractState> component, String message, byte flag, int devId) throws Exception
{
List<String> outputs = (flag == FLAG_GENERATE) ? new ArrayList<String>() : null;
if (message != null) LOG.info(message+"\n");
int i, total = 0, size = filenames.length;
StringBuilder build = null;
PrintStream out = null;
DEPTree tree;
String s;
for (i=0; i<size; i++)
{
reader.open(UTInput.createBufferedFileReader(filenames[i]));
switch (flag)
{
case FLAG_DECODE : out = new PrintStream(new BufferedOutputStream(new FileOutputStream(filenames[i]+".cnlp"))); break;
case FLAG_GENERATE: build = new StringBuilder();
}
while ((tree = reader.next()) != null)
{
component.process(tree, flag);
s = toString(tree)+"\n\n";
switch (flag)
{
case FLAG_DECODE : out.print(s); break;
case FLAG_GENERATE: build.append(s);
}
if (message != null && ++total%MAX_TREES == 0) LOG.info(".");
}
reader.close();
switch (flag)
{
case FLAG_DECODE : out.close(); break;
case FLAG_GENERATE: outputs.add(build.toString());
}
}
if (message != null) LOG.info("\n");
return outputs;
}
protected void train(AbstractOnlineStatisticalComponent<? extends AbstractState> component, NodeList eTrains, int boot)
{
StringModelAD[] models = component.getModels();
int modelSize = models.length;
AbstractAlgorithm algorithm;
StringModelAD model;
int i, nIterations;
Element eTrain;
for (i=0; i<modelSize; i++)
{
eTrain = (Element)eTrains.item(i);
model = models[i];
model.build(getLabelCutoff(eTrain), getFeatureCutoff(eTrain), getRandomSeed(eTrain), true);
model.printInfo(LOG);
nIterations = getNumberOfIterations(eTrain, boot);
algorithm = getAlgorithm(eTrain);
trainOnline(model, algorithm, nIterations);
}
}
private void trainOnline(StringModelAD model, AbstractAlgorithm algorithm, int nIterations)
{
int i;
for (i=0; i<nIterations; i++)
{
algorithm.train(model);
LOG.info(".");
}
LOG.info("\n");
}
@SuppressWarnings("unchecked")
protected double develop(String[] developFiles, JointReader reader, AbstractOnlineStatisticalComponent<? extends AbstractState> component, NodeList eTrains, double bootstrapScore, int boot, byte flag) throws Exception
{
StringModelAD[] models = component.getModels();
ObjectDoublePair<List<String>> output = null;
int modelSize = models.length;
AbstractAlgorithm algorithm;
StringModelAD model;
Element eTrain;
int i;
for (i=0; i<modelSize; i++)
{
eTrain = (Element)eTrains.item(i);
model = models[i];
model.build(getLabelCutoff(eTrain), getFeatureCutoff(eTrain), getRandomSeed(eTrain), true);
model.printInfo(LOG);
algorithm = getAlgorithm(eTrain);
output = developOnline(developFiles, reader, component, model, algorithm, bootstrapScore, flag);
}
if (flag == FLAG_GENERATE)
printOutput(developFiles, (List<String>)output.o, boot);
return output.d;
}
protected ObjectDoublePair<List<String>> developOnline(String[] developFiles, JointReader reader, AbstractOnlineStatisticalComponent<? extends AbstractState> component, StringModelAD model, AbstractAlgorithm algorithm, double bootstrapScore, byte flag) throws Exception
{
boolean prepareBootstrap = bootstrapScore > 0;
List<String> currOutput, bestOutput = null;
FloatArrayList bestWeights = null;
double currScore, bestScore = 0;
AbstractEval eval;
int iter;
for (iter=1; true; iter++)
{
algorithm.train(model);
currOutput = process(developFiles, reader, component, null, flag, -1);
eval = component.getEval();
currScore = eval.getAccuracies()[0];
LOG.info(String.format("%2d: %s\n", iter, eval.toString()));
eval.clear();
if (bestScore < currScore)
{
bestWeights = model.cloneWeights();
bestScore = currScore;
bestOutput = currOutput;
}
else break;
if (prepareBootstrap && bootstrapScore <= currScore) break;
}
model.setWeights(bestWeights);
return new ObjectDoublePair<List<String>>(bestOutput, bestScore);
}
protected ObjectDoublePair<List<String>> developBatch(String[] developFiles, JointReader reader, AbstractOnlineStatisticalComponent<? extends AbstractState> component, StringModelAD model, AbstractAlgorithm algorithm, byte flag) throws Exception
{
algorithm.train(model);
List<String> output = process(developFiles, reader, component, null, flag, -1);
AbstractEval eval = component.getEval();
double score = eval.getAccuracies()[0];
LOG.info(String.format("%s\n", eval.toString()));
eval.clear();
return new ObjectDoublePair<List<String>>(output, score);
}
protected void printOutput(String[] developFiles, List<String> output, int boot)
{
int i, size = developFiles.length;
PrintStream fout;
for (i=0; i<size; i++)
{
fout = UTOutput.createPrintBufferedFileStream(developFiles[i]+"."+boot);
fout.print(output.get(i));
fout.close();
}
}
// ================================== SUBCLASS ==================================
protected AbstractOnlineStatisticalComponent<? extends AbstractState> getCollector(JointFtrXml[] xmls, String[] trainFiles, JointReader reader, Element eConfig, int devId)
{
Element eMode = UTXml.getFirstElementByTagName(eConfig, MODE_POS);
int dfc = getDocumentFrequencyCutoff(eMode);
int dtc = getDocumentMaxTokenCount(eMode);
Set<String> sLsfs;
if (dtc <= 0) sLsfs = getLowerSimplifiedFormsByDocumentFrequencies(reader, trainFiles, devId, dfc);
else sLsfs = getLowerSimplifiedFormsByDocumentFrequencies(reader, trainFiles, devId, dfc, dtc);
return new OnlinePOSTagger(xmls, sLsfs);
}
protected AbstractOnlineStatisticalComponent<? extends AbstractState> getTrainer(JointFtrXml[] xmls, Object[] lexica)
{
return new OnlinePOSTagger(xmls, lexica);
}
protected AbstractOnlineStatisticalComponent<? extends AbstractState> getDecoder(ObjectInputStream in)
{
return new OnlinePOSTagger(in);
}
protected String toString(DEPTree tree)
{
return tree.toStringPOS();
}
private Set<String> getLowerSimplifiedFormsByDocumentFrequencies(JointReader reader, String[] filenames, int devId, int cutoff)
{
int i, j, len, size = filenames.length;
Set<String> set = Sets.newHashSet();
Prob1DMap map = new Prob1DMap();
DEPTree tree;
LOG.info(String.format("Collecting simplified-forms: cutoff = %d\n", cutoff));
for (i=0; i<size; i++)
{
if (i == devId) continue;
reader.open(UTInput.createBufferedFileReader(filenames[i]));
set.clear();
while ((tree = reader.next()) != null)
{
len = tree.size();
for (j=1; j<len; j++)
set.add(MPLib.getSimplifiedLowercaseWordForm(tree.get(j).form));
}
map.addAll(set);
reader.close();
LOG.info(".");
} LOG.info("\n");
return map.toSet(cutoff);
}
private Set<String> getLowerSimplifiedFormsByDocumentFrequencies(JointReader reader, String[] filenames, int devId, int cutoff, int maxCount)
{
int i, j, len, count = 0, size = filenames.length;
Set<String> set = Sets.newHashSet();
Prob1DMap map = new Prob1DMap();
DEPTree tree;
LOG.info(String.format("Collecting simplified-forms: cutoff = %d, max = %d\n", cutoff, maxCount));
for (i=0; i<size; i++)
{
if (i == devId) continue;
reader.open(UTInput.createBufferedFileReader(filenames[i]));
while ((tree = reader.next()) != null)
{
len = tree.size();
for (j=1; j<len; j++)
set.add(MPLib.getSimplifiedLowercaseWordForm(tree.get(j).form));
if ((count += len) >= maxCount)
{
map.addAll(set);
LOG.info(".");
set.clear();
count = 0;
}
}
reader.close();
} LOG.info("\n");
if (!set.isEmpty()) map.addAll(set);
return map.toSet(cutoff);
}
static public void main(String[] args)
{
new NLPDevelop(args);
}
}