/*
* RegressionGibbsEffectOperator.java
*
* Copyright (c) 2002-2015 Alexei Drummond, Andrew Rambaut and Marc Suchard
*
* This file is part of BEAST.
* See the NOTICE file distributed with this work for additional
* information regarding copyright ownership and licensing.
*
* BEAST is free software; you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as
* published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* BEAST is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with BEAST; if not, write to the
* Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
* Boston, MA 02110-1301 USA
*/
package dr.inference.operators;
import dr.inference.distribution.LinearRegression;
import dr.inference.distribution.MultivariateDistributionLikelihood;
import dr.inference.model.Parameter;
import dr.inferencexml.distribution.GeneralizedLinearModelParser;
import dr.math.distributions.MultivariateDistribution;
import dr.math.distributions.MultivariateNormalDistribution;
import dr.math.matrixAlgebra.SymmetricMatrix;
import dr.xml.*;
/**
* @author Marc Suchard
*/
public class RegressionGibbsEffectOperator extends SimpleMCMCOperator implements GibbsOperator {
public static final String GIBBS_OPERATOR = "regressionGibbsEffectOperator";
private LinearRegression linearModel;
private Parameter effect;
private Parameter indicators;
private boolean hasNoIndicators = true;
private MultivariateDistribution effectPrior;
private int dim;
private int effectNumber;
private int N;
private int numEffects;
private double[][] X;
private double[] mean = null;
private double[][] variance = null;
private double[][] precision = null;
public RegressionGibbsEffectOperator(LinearRegression linearModel, Parameter effect, Parameter indicators,
MultivariateDistributionLikelihood effectPrior) {
super();
this.linearModel = linearModel;
this.effect = effect;
this.indicators = indicators;
if (indicators != null) {
hasNoIndicators = false;
if (indicators.getDimension() != effect.getDimension())
throw new RuntimeException("Indicator and effect dimensions must match");
}
effectNumber = linearModel.getEffectNumber(effect);
this.effectPrior = effectPrior.getDistribution();
dim = effect.getDimension();
N = linearModel.getDependentVariable().getDimension();
numEffects = linearModel.getNumberOfFixedEffects();
X = linearModel.getX(effectNumber);
}
public int getStepCount() {
return 1;
}
public void computeForwardDensity(double[] outMean, double[][] outVariance, double[][] outPrecision) {
double[] W = linearModel.getTransformedDependentParameter();
double[] P = linearModel.getScale(); // outcome precision, fresh copy
for (int k = 0; k < numEffects; k++) {
if (k != effectNumber) {
double[] thisXBeta = linearModel.getXBeta(k);
for (int i = 0; i < N; i++)
W[i] -= thisXBeta[i];
}
}
double[] priorBetaMean = effectPrior.getMean();
double[][] priorBetaScale = effectPrior.getScaleMatrix();
double[][] XtP = new double[dim][N];
for (int j = 0; j < dim; j++) {
if (hasNoIndicators || indicators.getParameterValue(j) == 1) {
for (int i = 0; i < N; i++)
XtP[j][i] = X[i][j] * P[i];
} // else already filled with zeros
}
double[][] XtPX = new double[dim][dim];
for (int i = 0; i < dim; i++) {
if (hasNoIndicators || indicators.getParameterValue(i) == 1) {
for (int j = i; j < dim; j++) {// symmetric
if (hasNoIndicators || indicators.getParameterValue(j) == 1) {
for (int k = 0; k < N; k++)
XtPX[i][j] += XtP[i][k] * X[k][j];
XtPX[j][i] = XtPX[i][j]; // symmetric
}
}
}
}
double[][] XtPX_plus_P0 = new double[dim][dim];
for (int i = 0; i < dim; i++) {
for (int j = i; j < dim; j++) // symmetric
XtPX_plus_P0[j][i] = XtPX_plus_P0[i][j] = XtPX[i][j] + priorBetaScale[i][j];
}
double[] XtPW = new double[dim];
for (int i = 0; i < dim; i++) {
for (int j = 0; j < N; j++)
XtPW[i] += XtP[i][j] * W[j];
}
double[] P0Mean0 = new double[dim];
for (int i = 0; i < dim; i++) {
for (int j = 0; j < dim; j++)
P0Mean0[i] += priorBetaScale[i][j] * priorBetaMean[j];
}
double[] unscaledMean = new double[dim];
for (int i = 0; i < dim; i++)
unscaledMean[i] = P0Mean0[i] + XtPW[i];
double[][] variance = new SymmetricMatrix(XtPX_plus_P0).inverse().toComponents();
for (int i = 0; i < dim; i++) {
outMean[i] = 0.0;
for (int j = 0; j < dim; j++) {
outMean[i] += variance[i][j] * unscaledMean[j];
outVariance[i][j] = variance[i][j];
outPrecision[i][j] = XtPX_plus_P0[i][j];
}
}
}
public double[] getLastMean() { return mean; }
public double[][] getLastVariance() { return variance; }
public double[][] getLastPrecision() { return precision; }
public double doOperation() {
if (mean == null)
mean = new double[dim];
if (variance == null)
variance = new double[dim][dim];
if (precision == null)
precision = new double[dim][dim];
computeForwardDensity(mean,variance,precision);
double[] draw = MultivariateNormalDistribution.nextMultivariateNormalVariance(
mean, variance);
for (int i = 0; i < dim; i++)
effect.setParameterValue(i, draw[i]);
return 0;
}
public String getPerformanceSuggestion() {
return null;
}
public String getOperatorName() {
return GIBBS_OPERATOR;
}
public static dr.xml.XMLObjectParser PARSER = new dr.xml.AbstractXMLObjectParser() {
public String getParserName() {
return GIBBS_OPERATOR;
}
public Object parseXMLObject(XMLObject xo) throws XMLParseException {
double weight = xo.getDoubleAttribute(WEIGHT);
LinearRegression linearModel = (LinearRegression) xo.getChild(LinearRegression.class);
Parameter effect = (Parameter) xo.getChild(Parameter.class);
MultivariateDistributionLikelihood prior = (MultivariateDistributionLikelihood) xo.getChild(MultivariateDistributionLikelihood.class);
if (prior.getDistribution().getType().compareTo(MultivariateNormalDistribution.TYPE) != 0)
throw new XMLParseException("Only a multivariate normal prior is conjugate");
XMLObject cxo = xo.getChild(GeneralizedLinearModelParser.INDICATOR);
Parameter indicators = null;
if (cxo != null) {
indicators = (Parameter) cxo.getChild(Parameter.class);
}
RegressionGibbsEffectOperator operator = new RegressionGibbsEffectOperator(linearModel, effect, indicators, prior);
operator.setWeight(weight);
return operator;
}
//************************************************************************
// AbstractXMLObjectParser implementation
//************************************************************************
public String getParserDescription() {
return "This element returns a multivariate Gibbs operator on an internal node trait.";
}
public Class getReturnType() {
return MCMCOperator.class;
}
public XMLSyntaxRule[] getSyntaxRules() {
return rules;
}
private XMLSyntaxRule[] rules = new XMLSyntaxRule[]{
AttributeRule.newDoubleRule(WEIGHT),
new ElementRule(Parameter.class),
new ElementRule(MultivariateDistributionLikelihood.class),
new ElementRule(LinearRegression.class),
new ElementRule(GeneralizedLinearModelParser.INDICATOR,
new XMLSyntaxRule[] {
new ElementRule(Parameter.class)
},true)
};
};
}