com.github.steveash.jg2p.align.AlignerViterbi Maven / Gradle / Ivy
The newest version!
/*
* Copyright 2014 Steve Ash
*
* 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.github.steveash.jg2p.align;
import com.google.common.collect.Lists;
import com.google.common.collect.Ordering;
import com.google.common.math.DoubleMath;
import com.github.steveash.jg2p.Grams;
import com.github.steveash.jg2p.Word;
import com.github.steveash.jg2p.syll.SWord;
import java.util.Collections;
import java.util.List;
/**
* Implementation of the viterbi algorithm to walk the k-most probable sequences through the inferred prob table
* @author Steve Ash
*/
public class AlignerViterbi {
public AlignerViterbi(GramOptions opts, ProbTable probs) {
this.opts = opts;
this.probs = probs;
this.penalizer = opts.makePenalizer();
}
private final GramOptions opts;
private final ProbTable probs;
private final Penalizer penalizer;
public List align(Word x, Word y, int bestPathCount) {
PathXYTable t = new PathXYTable(x.unigramCount() + 1, y.unigramCount() + 1, bestPathCount);
t.offer(0, 0, t.make(0, -1, -1, -1));
for (int xx = 0; xx < x.unigramCount() + 1; xx++) {
for (int yy = 0; yy < y.unigramCount() + 1; yy++) {
if (xx > 0 && opts.isIncludeXEpsilons()) {
for (int i = 1; (i <= opts.getMaxXGram()) && (xx - i >= 0); i++) {
String xGram = x.gram(xx - i, i);
double score = DoubleMath.log2(penalizer.penalize(xGram, Grams.EPSILON, probs.prob(xGram, Grams.EPSILON))); // what was this * i business
t.extendPath(xx, yy, xx - i, yy, PathXYTable.Entry.sample(score, i, 0));
}
}
if (yy > 0 && opts.isIncludeEpsilonYs()) {
for (int j = 1; (j <= opts.getMaxYGram()) && (yy - j >= 0); j++) {
String yGram = y.gram(yy - j, j);
double score = DoubleMath.log2(penalizer.penalize(Grams.EPSILON, yGram, probs.prob(Grams.EPSILON, yGram))); // * j;
t.extendPath(xx, yy, xx, yy - j, PathXYTable.Entry.sample(score, 0, j));
}
}
if (xx > 0 && yy > 0) {
for (int i = 1; (i <= opts.getMaxXGram()) && (xx - i >= 0); i++) {
for (int j = 1; (j <= opts.getMaxYGram()) && (yy - j >= 0); j++) {
String xGram = x.gram(xx - i, i);
String yGram = y.gram(yy - j, j);
double score = DoubleMath.log2(penalizer.penalize(xGram, yGram, probs.prob(xGram, yGram))); // * Math.max(i, j);
t.extendPath(xx, yy, xx - i, yy - j, PathXYTable.Entry.sample(score, i, j));
}
}
}
}
}
return createAlignments(x, y, t, bestPathCount);
}
private List createAlignments(Word x, Word y, PathXYTable t, int bestPathCount) {
List results = Lists.newArrayListWithCapacity(bestPathCount);
Iterable lastEntries = t.get(x.unigramCount(), y.unigramCount());
for (PathXYTable.Entry lastEntry : lastEntries) {
if (lastEntry.score < ProbTable.minLogProb) continue;
results.add(decodePathFrom(x, y, t, lastEntry));
}
Collections.sort(results, Ordering.natural().reverse());
return results;
}
private Alignment decodePathFrom(Word x, Word y, PathXYTable t, PathXYTable.Entry entry) {
int xx = x.unigramCount();
int yy = y.unigramCount();
Alignment a = new Alignment(x, entry.score);
while (xx > 0 || yy > 0) {
String xGram = x.gram(xx - entry.xBackRef, entry.xBackRef);
String yGram = y.gram(yy - entry.yBackRef, entry.yBackRef);
a.append(xGram, yGram);
xx -= entry.xBackRef;
yy -= entry.yBackRef;
entry = t.get(xx, yy, entry.pathBackRef);
}
Alignment finish = a.finish();
if (y instanceof SWord) {
finish = finish.withSyllWord((SWord) y);
}
return finish;
}
}