List of usage examples for org.apache.commons.math3.analysis.function Exp Exp
Exp
From source file:edu.stanford.cfuller.imageanalysistools.clustering.GaussianLikelihoodObjectiveFunction.java
/** * Evaluates the function with the specified parameters. * * The parameters describe a set of gaussian generators (which are the Clusters). * * @param parameters A RealVector containing the values of all the parameters of each Gaussian, ordered so that all the parameters of a single gaussian are together, then the next gaussian, etc. * @return The negative log likelihood of having observed the ClusterObjects at their locations, given the parameters describing the Gaussian clusters. *///from w w w . j av a 2 s . c o m public double evaluate(RealVector parameters) { int nClusters = parameters.getDimension() / nParametersEach; //java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("nClusters: " + nClusters + " abdMatrices_size: " + abdMatrices.size() + " det_dim: " + det.getDimension()); if (det.getDimension() != nClusters) { clusterProbs = new Array2DRowRealMatrix(this.objects.size(), nClusters); det = new ArrayRealVector(nClusters); pk = new ArrayRealVector(nClusters); if (abdMatrices.size() < nClusters) { int originalSize = abdMatrices.size(); for (int i = 0; i < nClusters - originalSize; i++) { abdMatrices.add(new Array2DRowRealMatrix(numDim, numDim)); } } else { abdMatrices.setSize(nClusters); } } pk.mapMultiplyToSelf(0.0); //java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("nClusters: " + nClusters + " abdMatrices_size: " + abdMatrices.size() + " det_dim: " + det.getDimension()); for (int i = 0; i < nClusters; i++) { /* double ct = Math.cos(parameters.getEntry(nParametersEach*i+3)); double st = Math.sin(parameters.getEntry(nParametersEach*i+3)); double sin2t = Math.sin(2*parameters.getEntry(nParametersEach*i+3)); double a = (ct*ct/(2*parameters.getEntry(nParametersEach*i+2)) + st*st/(2*parameters.getEntry(nParametersEach*i+4))); double b = (sin2t/(4*parameters.getEntry(nParametersEach*i+4)) - sin2t/(4*parameters.getEntry(nParametersEach*i+2))); double d = (st*st/(2*parameters.getEntry(nParametersEach*i+2)) + ct*ct/(2*parameters.getEntry(nParametersEach*i+4))); */ double a = parameters.getEntry(nParametersEach * i + 2); double d = parameters.getEntry(nParametersEach * i + 4); double b = Math.sqrt(a * d) * parameters.getEntry(nParametersEach * i + 3); abdMatrices.get(i).setEntry(0, 0, a); abdMatrices.get(i).setEntry(1, 0, b); abdMatrices.get(i).setEntry(0, 1, b); abdMatrices.get(i).setEntry(1, 1, d); LUDecomposition abdLU = (new LUDecomposition(abdMatrices.get(i))); det.setEntry(i, (abdLU).getDeterminant()); //det.setEntry(i, a*d-b*b); try { abdMatrices.set(i, abdLU.getSolver().getInverse()); } catch (org.apache.commons.math3.linear.SingularMatrixException e) { return Double.MAX_VALUE; } } for (int n = 0; n < this.objects.size(); n++) { ClusterObject c = this.objects.get(n); double max = -1.0 * Double.MAX_VALUE; int maxIndex = 0; for (int k = 0; k < nClusters; k++) { mean.setEntry(0, c.getCentroid().getX() - parameters.getEntry(nParametersEach * k)); mean.setEntry(1, c.getCentroid().getY() - parameters.getEntry(nParametersEach * k + 1)); x = abdMatrices.get(k).operate(mean); double dot = x.dotProduct(mean); // java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("k, n: " + k + ", " + this.objects.size()); // java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("parameters: " + parameters.toString()); // java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("abd matrix: " + abdMatrices.get(k).toString()); // java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("det: " + det.getEntry(k)); // java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("mean: " + mean.toString()); // java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("dot: " + dot); double logN = negLog2PI - 0.5 * Math.log(det.getEntry(k)) - 0.5 * dot; // java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("logN: " + logN); clusterProbs.setEntry(n, k, logN); if (logN > max) { max = logN; maxIndex = k; } if (Double.isInfinite(logN) || Double.isNaN(logN)) { return Double.MAX_VALUE; } } c.setMostProbableCluster(maxIndex); } for (int k = 0; k < nClusters; k++) { double tempMax = -1.0 * Double.MAX_VALUE; for (int n = 0; n < this.objects.size(); n++) { if (clusterProbs.getEntry(n, k) > tempMax) tempMax = clusterProbs.getEntry(n, k); } pk.setEntry(k, tempMax + Math.log( clusterProbs.getColumnVector(k).mapSubtract(tempMax).mapToSelf(new Exp()).getL1Norm()) - Math.log(this.objects.size())); } double pkMax = -1.0 * Double.MAX_VALUE; for (int k = 0; k < nClusters; k++) { if (pk.getEntry(k) > pkMax) pkMax = pk.getEntry(k); } double logSumPk = pkMax + Math.log(pk.mapSubtract(pkMax).mapToSelf(new Exp()).getL1Norm()); pk.mapSubtractToSelf(logSumPk); //java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("pk: " + pk.toString()); double L = 0; for (int n = 0; n < this.objects.size(); n++) { RealVector toSum = clusterProbs.getRowVector(n).add(pk); double tempMax = -1.0 * Double.MAX_VALUE; for (int k = 0; k < nClusters; k++) { if (toSum.getEntry(k) > tempMax) tempMax = toSum.getEntry(k); } double pn = tempMax + Math.log(toSum.mapSubtract(tempMax).mapToSelf(new Exp()).getL1Norm()); //java.util.logging.Logger.getLogger("edu.stanford.cfuller.imageanalysistools").info("pn: " + pn); L += pn; } return -1.0 * L; }
From source file:org.deidentifier.arx.criteria.EDDifferentialPrivacy.java
/** * Calculates beta_max/*from www . j av a 2 s . co m*/ * @param epsilon * @return */ private double calculateBeta(double epsilon) { return 1.0d - (new Exp()).value(-1.0d * epsilon); }
From source file:org.deidentifier.arx.criteria.EDDifferentialPrivacy.java
/** * Calculates c_n/*from w w w .j a v a 2s .c o m*/ * @param n * @param epsilon * @param beta * @return */ private double calculateC(int n, double epsilon, double beta) { double gamma = calculateGamma(epsilon, beta); return (new Exp()).value(-1.0d * n * (gamma * (new Log()).value(gamma / beta) - (gamma - beta))); }
From source file:org.deidentifier.arx.criteria.EDDifferentialPrivacy.java
/** * Calculates gamma// w w w.jav a 2 s . c om * @param epsilon * @param beta * @return */ private double calculateGamma(double epsilon, double beta) { double power = (new Exp()).value(epsilon); return (power - 1.0d + beta) / power; }