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.math.statistics.normalisation;
031
032import org.openimaj.math.util.DoubleArrayStatsUtils;
033import org.openimaj.util.array.ArrayUtils;
034
035/**
036 * Subtract the mean of each example vector from itself and divide by the
037 * standard deviation to normalise the vector such that it has unit variance. A
038 * regularisation term can be optionally included in the divisor.
039 * <p>
040 * Only use if the data is stationary (i.e., the statistics for each data
041 * dimension follow the same distribution).
042 *
043 * @author Jonathon Hare (jsh2@ecs.soton.ac.uk)
044 */
045public class PerExampleMeanCenterVar implements Normaliser {
046        double eps = 10.0 / 255.0;
047
048        /**
049         * Construct with the given variance regularisation term. Setting to zero
050         * disables the regulariser.
051         *
052         * @param eps
053         *            the variance normalisation regularizer (each dimension is
054         *            divided by sqrt(var + eps).
055         */
056        public PerExampleMeanCenterVar(double eps) {
057                this.eps = eps;
058        }
059
060        @Override
061        public double[] normalise(double[] vector) {
062                final double mean = DoubleArrayStatsUtils.mean(vector);
063                final double var = DoubleArrayStatsUtils.var(vector);
064                vector = ArrayUtils.subtract(vector, mean);
065                vector = ArrayUtils.divide(vector, Math.sqrt(var + eps));
066
067                return vector;
068        }
069
070        @Override
071        public double[][] normalise(double[][] data) {
072                final double[][] out = new double[data.length][];
073                for (int c = 0; c < out.length; c++)
074                        out[c] = normalise(data[c]);
075                return out;
076        }
077}