001/**
002 * Copyright (c) 2011, The University of Southampton and the individual contributors.
003 * All rights reserved.
004 *
005 * Redistribution and use in source and binary forms, with or without modification,
006 * are permitted provided that the following conditions are met:
007 *
008 *   *  Redistributions of source code must retain the above copyright notice,
009 *      this list of conditions and the following disclaimer.
010 *
011 *   *  Redistributions in binary form must reproduce the above copyright notice,
012 *      this list of conditions and the following disclaimer in the documentation
013 *      and/or other materials provided with the distribution.
014 *
015 *   *  Neither the name of the University of Southampton nor the names of its
016 *      contributors may be used to endorse or promote products derived from this
017 *      software without specific prior written permission.
018 *
019 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
020 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
021 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
022 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
023 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
024 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
025 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
026 * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
027 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
028 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
029 */
030package org.openimaj.pgm.vb.lda.mle;
031
032import org.openimaj.pgm.util.Document;
033import org.openimaj.util.array.SparseIntArray.Entry;
034
035/**
036 * The state of the E step of the MLE LDA
037 * @author Sina Samangooei (ss@ecs.soton.ac.uk)
038 *
039 */
040public class LDAVariationlState{
041        /**
042         * the n'th unique word in a document's probability for each topic
043         */
044        public double[][] phi;
045        /**
046         * Useful for calculating the sumphi for a given document
047         */
048        public double[] oldphi;
049        /**
050         * the dirichlet parameter for the topic multinomials
051         */
052        public double[] varGamma;
053        
054        /**
055         * The liklihood of the current topic sufficient statistics and variational parameters
056         */
057        public double likelihood; 
058        /**
059         * The old liklihood
060         */
061        public double oldLikelihood;
062        /**
063         * The current LDAModel (i.e. the current sufficient statistics
064         */
065        public LDAModel state;
066        /**
067         * Holds the first derivative of the gamma 
068         */
069        public double[] digamma;
070        
071        
072        int iteration;
073        
074
075        /**
076         * The variational state holds phi and gamma states as well as 
077         * information for convergence of the E step.
078         * @param state
079         */
080        public LDAVariationlState(LDAModel state) {
081                this.oldphi = new double[state.ntopics];
082                this.varGamma = new double[state.ntopics];
083                this.digamma = new double[state.ntopics];
084                this.state = state;
085        }
086
087        /**
088         * initialises the phi and sets everything to 0
089         * @param doc
090         */
091        public void prepare(Document doc){
092                this.phi = new double[doc.countUniqueWords()][state.ntopics];
093                likelihood = 0;
094                oldLikelihood = Double.NEGATIVE_INFINITY;
095                for (int topici = 0; topici < phi.length; topici++) {
096                        varGamma[topici] = this.state.alpha;
097                        digamma[topici] = 0; // used to calculate likelihood
098                        int wordi = 0;
099                        for (Entry wordCount : doc.getVector().entries()) {
100                                phi[wordi][topici] = 1f/this.state.ntopics;
101                                varGamma[topici] += (double)wordCount.value / this.state.ntopics;
102                                wordi++;
103                        }
104                }
105                this.iteration = 0;
106        }
107}