edu.berkeley.nlp.crf.Inference Maven / Gradle / Ivy

Go to download

Show more of this group Show more artifacts with this name
Show all versions of berkeleyparser Show documentation

The Berkeley parser analyzes the grammatical structure of natural language using probabilistic context-free grammars (PCFGs).

The newest version!

package edu.berkeley.nlp.crf;

import java.io.Serializable;

import edu.berkeley.nlp.classify.Encoding;
import edu.berkeley.nlp.classify.FeatureExtractor;
import edu.berkeley.nlp.math.DoubleArrays;
import edu.berkeley.nlp.math.DoubleMatrices;
import edu.berkeley.nlp.util.ArrayUtil;
import edu.berkeley.nlp.util.Pair;
import edu.berkeley.nlp.util.PriorityQueue;

public class Inference implements Serializable {
	/**
	 * 
	 */
	private static final long serialVersionUID = 1948395432745606240L;
	
	private final Encoding encoding;
	private final ScoreCalculator scoreCalculator;
	
	public Inference(Encoding encoding,
			FeatureExtractor vertexExtractor, FeatureExtractor edgeExtractor) {
		this.encoding = encoding;
		this.scoreCalculator = new ScoreCalculator(encoding, vertexExtractor, edgeExtractor);
	}
	
	public double[][] getAlphas(InstanceSequence sequence, double[] w) {
		int n = sequence.getSequenceLength();
		double[][] alpha = new double[n][];
		alpha[0] = scoreCalculator.getVertexScores(sequence, 0, w);
		for (int i=1; i sequence, double[] w) {
		int n = sequence.getSequenceLength();
		double[][] beta = new double[n][];
		beta[n-1] = DoubleArrays.constantArray(1.0, encoding.getNumLabels());
		for (int i=n-2; i>=0; i--) {
			double[][] scoreMatrix = scoreCalculator.getScoreMatrix(sequence, i+1, w);
			beta[i] = DoubleMatrices.product(scoreMatrix, beta[i+1]);
		}
		return beta;
	}
	
	public Pair getKBestChartAndBacktrace(InstanceSequence sequence, double[] w, int k) {
		int n = sequence.getSequenceLength();
		int numLabels = encoding.getNumLabels();
		int[][][][] bestLabels = new int[n][numLabels][][];
		double[][][] bestScores = new double[n][numLabels][];
		double[] startScores = scoreCalculator.getLinearVertexScores(sequence, 0, w);
		for (int l=0; l> pq = new PriorityQueue>();
				for (int pl=0; pl backtrace = pq.next();
					bestLabels[i][l][c][0] = backtrace.getFirst();
					bestLabels[i][l][c][1] = backtrace.getSecond();
				}
			}
		}
		return Pair.makePair(bestLabels, bestScores);
	}
	
	public double[][] getVertexPosteriors(double[][] alpha, double[][] beta) {
		double[][] p = new double[alpha.length][encoding.getNumLabels()];
		for (int i=0; i sequence, double[] w, double[][] alpha, double[][] beta) {
		int numLabels = encoding.getNumLabels();
		int n = sequence.getSequenceLength();
		double[][][] p = new double[n][numLabels][numLabels];
		for (int i=1; i