jidt/java/source/infodynamics/measures/continuous/gaussian/OInfoCalculatorGaussian.java

135 lines
4.9 KiB
Java

/*
* Java Information Dynamics Toolkit (JIDT)
* Copyright (C) 2017, Joseph T. Lizier, Ipek Oezdemir and Pedro Mediano
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program 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 General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
package infodynamics.measures.continuous.gaussian;
import infodynamics.utils.MatrixUtils;
/**
* <p>Computes the differential O-information of a given multivariate
* <code>double[][]</code> set of
* observations (extending {@link MultiVariateInfoMeasureCalculatorGaussian}),
* assuming that the probability distribution function for these observations is
* a multivariate Gaussian distribution.</p>
*
* <p>Usage is as per the paradigm outlined for {@link MultiVariateInfoMeasureCalculatorCommon}.
* </p>
*
* <p><b>References:</b><br/>
* <ul>
* <li>Rosas, F., Mediano, P., Gastpar, M, Jensen, H.,
* <a href="http://dx.doi.org/10.1103/PhysRevE.100.032305">"Quantifying high-order
* interdependencies via multivariate extensions of the mutual information"</a>,
* Physical Review E 100, (2019) 032305.</li>
* </ul>
*
* @author Pedro A.M. Mediano (<a href="pmediano at pm.me">email</a>,
* <a href="http://www.doc.ic.ac.uk/~pam213">www</a>)
*/
public class OInfoCalculatorGaussian
extends MultiVariateInfoMeasureCalculatorGaussian {
/**
* Constructor.
*/
public OInfoCalculatorGaussian() {
// Nothing to do
}
/**
* {@inheritDoc}
*
* @return the average O-info in nats (not bits!)
* @throws Exception if not sufficient data have been provided, or if the
* supplied covariance matrix is invalid.
*/
public double computeAverageLocalOfObservations() throws Exception {
if (covariance == null) {
throw new Exception("Cannot calculate O-Info without having " +
"a covariance either supplied or computed via setObservations()");
}
if (!isComputed) {
double oinfo = (dimensions - 2)*Math.log(MatrixUtils.determinantSymmPosDefMatrix(covariance));
for (int i = 0; i < dimensions; i++) {
int[] idx = allExcept(i, dimensions);
double[][] marginal_cov = MatrixUtils.selectRowsAndColumns(covariance, idx, idx);
oinfo += Math.log(covariance[i][i]) - Math.log(MatrixUtils.determinantSymmPosDefMatrix(marginal_cov));
}
// This "0.5" comes from the entropy formula for Gaussians: h = 0.5*logdet(2*pi*e*Sigma)
lastAverage = 0.5*oinfo;;
isComputed = true;
}
return lastAverage;
}
/**
* {@inheritDoc}
*
* @return the "time-series" of local O-info values in nats (not bits!)
* for the supplied states.
* @throws Exception if not sufficient data have been provided, or if the
* supplied covariance matrix is invalid.
*/
public double[] computeLocalUsingPreviousObservations(double[][] states) throws Exception {
if ((means == null) || (covariance == null)) {
throw new Exception("Cannot compute local values without having means " +
"and covariance either supplied or computed via setObservations()");
}
EntropyCalculatorMultiVariateGaussian hCalc = new EntropyCalculatorMultiVariateGaussian();
hCalc.initialise(dimensions);
hCalc.setCovarianceAndMeans(covariance, means);
double[] localValues = MatrixUtils.multiply(hCalc.computeLocalUsingPreviousObservations(states), dimensions - 2);
for (int i = 0; i < dimensions; i++) {
int[] idx = allExcept(i, dimensions);
// Local entropy of this variable (i) only
double[][] this_cov = MatrixUtils.selectRowsAndColumns(covariance, i, 1, i, 1);
double[] this_means = MatrixUtils.select(means, i, 1);
double[][] this_state = MatrixUtils.selectColumns(states, i, 1);
hCalc.initialise(1);
hCalc.setCovarianceAndMeans(this_cov, this_means);
double[] thisLocals = hCalc.computeLocalUsingPreviousObservations(this_state);
// Local entropy of the rest of the variables (0, ... i-1, i+1, ... D)
double[][] rest_cov = MatrixUtils.selectRowsAndColumns(covariance, idx, idx);
double[] rest_means = MatrixUtils.select(means, idx);
double[][] rest_state = MatrixUtils.selectColumns(states, idx);
hCalc.initialise(dimensions - 1);
hCalc.setCovarianceAndMeans(rest_cov, rest_means);
double[] restLocals = hCalc.computeLocalUsingPreviousObservations(rest_state);
localValues = MatrixUtils.add(localValues, MatrixUtils.subtract(thisLocals, restLocals));
}
return localValues;
}
}