com.gengoai.apollo.ml.model.sequence.MalletCrf Maven / Gradle / Ivy
/*
* 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 com.gengoai.apollo.ml.model.sequence;
import cc.mallet.fst.*;
import cc.mallet.optimize.LimitedMemoryBFGS;
import cc.mallet.optimize.Optimizable;
import cc.mallet.pipe.SerialPipes;
import cc.mallet.pipe.TokenSequence2FeatureVectorSequence;
import cc.mallet.types.*;
import cc.mallet.util.MalletLogger;
import com.gengoai.ParameterDef;
import com.gengoai.apollo.ml.DataSet;
import com.gengoai.apollo.ml.Datum;
import com.gengoai.apollo.ml.model.LabelType;
import com.gengoai.apollo.ml.model.Params;
import com.gengoai.apollo.ml.model.SingleSourceFitParameters;
import com.gengoai.apollo.ml.model.SingleSourceModel;
import com.gengoai.apollo.ml.observation.Observation;
import com.gengoai.apollo.ml.observation.Variable;
import com.gengoai.apollo.ml.observation.VariableSequence;
import com.gengoai.conversion.Cast;
import lombok.NonNull;
import java.util.Arrays;
import java.util.function.Consumer;
import java.util.logging.Level;
import java.util.regex.Pattern;
import static com.gengoai.apollo.ml.model.sequence.Order.FIRST;
import static com.gengoai.collection.Arrays2.arrayOfInt;
import static com.gengoai.function.Functional.with;
/**
* A wrapper around Mallet's CRF implementation
*
* @author David B. Bracewell
*/
public class MalletCrf extends SingleSourceModel {
private static final long serialVersionUID = 1L;
public static final ParameterDef FULLY_CONNECTED = ParameterDef.boolParam("fullyConnected");
public static final ParameterDef ORDER = ParameterDef.param("order", Order.class);
public static final ParameterDef START_STATE = ParameterDef.strParam("startState");
public static final ParameterDef THREADS = ParameterDef.intParam("numThreads");
private SerialPipes pipes;
private CRF model;
private String startState;
/**
* Instantiates a new MalletCrf with default parameters.
*/
public MalletCrf() {
super(new Parameters());
}
/**
* Instantiates a new MalletCrf with the given parameters.
*
* @param parameters the parameters
*/
public MalletCrf(@NonNull Parameters parameters) {
super(parameters);
}
/**
* Instantiates a new MalletCrf with the given parameter updater.
*
* @param updater the updater
*/
public MalletCrf(@NonNull Consumer updater) {
super(with(new Parameters(), updater));
}
@Override
public void estimate(DataSet preprocessed) {
if(parameters.verbose.value()) {
MalletLogger.getLogger(ThreadedOptimizable.class.getName())
.setLevel(Level.INFO);
MalletLogger.getLogger(CRFTrainerByValueGradients.class.getName())
.setLevel(Level.INFO);
MalletLogger.getLogger(CRF.class.getName())
.setLevel(Level.INFO);
MalletLogger.getLogger(CRFOptimizableByBatchLabelLikelihood.class.getName())
.setLevel(Level.INFO);
MalletLogger.getLogger(LimitedMemoryBFGS.class.getName())
.setLevel(Level.INFO);
} else {
MalletLogger.getLogger(ThreadedOptimizable.class.getName())
.setLevel(Level.OFF);
MalletLogger.getLogger(CRFTrainerByValueGradients.class.getName())
.setLevel(Level.OFF);
MalletLogger.getLogger(CRF.class.getName())
.setLevel(Level.OFF);
MalletLogger.getLogger(CRFOptimizableByBatchLabelLikelihood.class.getName())
.setLevel(Level.OFF);
MalletLogger.getLogger(LimitedMemoryBFGS.class.getName())
.setLevel(Level.OFF);
}
Alphabet dataAlphabet = new Alphabet();
pipes = new SerialPipes(Arrays.asList(new SequenceToTokenSequence(),
new TokenSequence2FeatureVectorSequence(dataAlphabet, false, true)));
pipes.setDataAlphabet(dataAlphabet);
pipes.setTargetAlphabet(new LabelAlphabet());
InstanceList trainingData = new InstanceList(pipes);
for(Datum datum : preprocessed) {
com.gengoai.apollo.ml.observation.Sequence x = datum.get(parameters.input.value()).asSequence();
com.gengoai.apollo.ml.observation.Sequence y = datum.get(parameters.output.value()).asSequence();
Label[] target = new Label[x.size()];
LabelAlphabet labelAlphabet = Cast.as(trainingData.getTargetAlphabet());
for(int j = 0; j < target.length; j++) {
target[j] = labelAlphabet.lookupLabel(y.get(j).asVariable().getName(), true);
}
trainingData.addThruPipe(new Instance(x, new LabelSequence(target), null, null));
}
model = new CRF(pipes, null);
int[] order = {};
switch(parameters.order.value()) {
case FIRST:
order = arrayOfInt(1);
break;
case SECOND:
order = arrayOfInt(1, 2);
break;
case THIRD:
order = arrayOfInt(1, 2, 3);
break;
}
MalletSequenceValidator sv = Cast.as(parameters.validator.value() instanceof MalletSequenceValidator
? parameters.validator.value()
: null);
Pattern allowed = sv == null
? null
: sv.getAllowed();
Pattern forbidden = sv == null
? null
: sv.getForbidden();
model.addOrderNStates(trainingData,
order,
null,
parameters.startState.value(),
forbidden,
allowed,
parameters.fullyConnected.value());
this.startState = parameters.startState.value();
model.setWeightsDimensionAsIn(trainingData, false);
CRFOptimizableByBatchLabelLikelihood batchOptLabel = new CRFOptimizableByBatchLabelLikelihood(model,
trainingData,
parameters.numberOfThreads
.value());
ThreadedOptimizable optLabel = new ThreadedOptimizable(batchOptLabel,
trainingData,
model.getParameters().getNumFactors(),
new CRFCacheStaleIndicator(model));
Optimizable.ByGradientValue[] opts = {optLabel};
CRFTrainerByValueGradients crfTrainer = new CRFTrainerByValueGradients(model, opts);
crfTrainer.setMaxResets(0);
crfTrainer.train(trainingData, parameters.maxIterations.value());
optLabel.shutdown();
}
@Override
public Parameters getFitParameters() {
return new Parameters();
}
@Override
public LabelType getLabelType(@NonNull String name) {
if(parameters.output.value().equals(name)) {
return LabelType.Sequence;
}
throw new IllegalArgumentException("'" + name + "' is not a valid output for this model.");
}
@Override
protected Observation transform(@NonNull Observation observation) {
int length = observation.asSequence().size();
Sequence sequence = Cast.as(model.getInputPipe()
.instanceFrom(new Instance(observation, null, null, null)).getData());
Sequence bestOutput = model.transduce(sequence);
SumLattice lattice = new SumLatticeDefault(model, sequence, true);
Transducer.State sj = model.getState(startState);
VariableSequence labeling = new VariableSequence();
for(int i = 0; i < sequence.size(); i++) {
Transducer.State si = model.getState((String) bestOutput.get(i));
String label = (String) bestOutput.get(i);
double pS = lattice.getGammaProbability(i, si);
double PSjSi = lattice.getXiProbability(i, sj, si);
double score = Math.max(pS, PSjSi);
sj = si;
labeling.add(Variable.real(label, score));
}
return labeling;
}
@Override
protected void updateMetadata(@NonNull DataSet data) {
data.updateMetadata(parameters.output.value(), m -> {
m.setEncoder(null);
m.setType(VariableSequence.class);
m.setDimension(-1);
});
}
/**
* MalletCrf Fit Parameters
*/
public static class Parameters extends SingleSourceFitParameters {
private static final long serialVersionUID = 1L;
/**
* The sequence validator to user during inference (default {@link SequenceValidator#ALWAYS_TRUE})
*/
public final Parameter validator = parameter(Params.Sequence.validator,
SequenceValidator.ALWAYS_TRUE);
/**
* The number of threads to use for training (default 20)
*/
public final Parameter numberOfThreads = parameter(THREADS, 20);
/**
* The order of the CRF (default {@link Order#FIRST})
*/
public final Parameter order = parameter(ORDER, FIRST);
/**
* The maximum number of iterations to run for (default 250).
*/
public final Parameter maxIterations = parameter(Params.Optimizable.maxIterations, 250);
/**
* Parameter denoting whether the generated graphical model is fully connected (default true).
*/
public final Parameter fullyConnected = parameter(FULLY_CONNECTED, true);
/**
* Parameter denoting the start state (default O).
*/
public final Parameter startState = parameter(START_STATE, "O");
}
}//END OF MalletCRF
© 2015 - 2025 Weber Informatics LLC | Privacy Policy