001/*
002 * Licensed to the Apache Software Foundation (ASF) under one or more
003 * contributor license agreements.  See the NOTICE file distributed with
004 * this work for additional information regarding copyright ownership.
005 * The ASF licenses this file to You under the Apache License, Version 2.0
006 * (the "License"); you may not use this file except in compliance with
007 * the License.  You may obtain a copy of the License at
008 *
009 *      http://www.apache.org/licenses/LICENSE-2.0
010 *
011 * Unless required by applicable law or agreed to in writing, software
012 * distributed under the License is distributed on an "AS IS" BASIS,
013 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
014 * See the License for the specific language governing permissions and
015 * limitations under the License.
016 */
017package org.apache.commons.math3.stat.correlation;
018
019import org.apache.commons.math3.exception.DimensionMismatchException;
020import org.apache.commons.math3.linear.BlockRealMatrix;
021import org.apache.commons.math3.linear.MatrixUtils;
022import org.apache.commons.math3.linear.RealMatrix;
023import org.apache.commons.math3.util.FastMath;
024import org.apache.commons.math3.util.Pair;
025
026import java.util.Arrays;
027import java.util.Comparator;
028
029/**
030 * Implementation of Kendall's Tau-b rank correlation</a>.
031 * <p>
032 * A pair of observations (x<sub>1</sub>, y<sub>1</sub>) and
033 * (x<sub>2</sub>, y<sub>2</sub>) are considered <i>concordant</i> if
034 * x<sub>1</sub> &lt; x<sub>2</sub> and y<sub>1</sub> &lt; y<sub>2</sub>
035 * or x<sub>2</sub> &lt; x<sub>1</sub> and y<sub>2</sub> &lt; y<sub>1</sub>.
036 * The pair is <i>discordant</i> if x<sub>1</sub> &lt; x<sub>2</sub> and
037 * y<sub>2</sub> &lt; y<sub>1</sub> or x<sub>2</sub> &lt; x<sub>1</sub> and
038 * y<sub>1</sub> &lt; y<sub>2</sub>.  If either x<sub>1</sub> = x<sub>2</sub>
039 * or y<sub>1</sub> = y<sub>2</sub>, the pair is neither concordant nor
040 * discordant.
041 * <p>
042 * Kendall's Tau-b is defined as:
043 * <pre>
044 * tau<sub>b</sub> = (n<sub>c</sub> - n<sub>d</sub>) / sqrt((n<sub>0</sub> - n<sub>1</sub>) * (n<sub>0</sub> - n<sub>2</sub>))
045 * </pre>
046 * <p>
047 * where:
048 * <ul>
049 *     <li>n<sub>0</sub> = n * (n - 1) / 2</li>
050 *     <li>n<sub>c</sub> = Number of concordant pairs</li>
051 *     <li>n<sub>d</sub> = Number of discordant pairs</li>
052 *     <li>n<sub>1</sub> = sum of t<sub>i</sub> * (t<sub>i</sub> - 1) / 2 for all i</li>
053 *     <li>n<sub>2</sub> = sum of u<sub>j</sub> * (u<sub>j</sub> - 1) / 2 for all j</li>
054 *     <li>t<sub>i</sub> = Number of tied values in the i<sup>th</sup> group of ties in x</li>
055 *     <li>u<sub>j</sub> = Number of tied values in the j<sup>th</sup> group of ties in y</li>
056 * </ul>
057 * <p>
058 * This implementation uses the O(n log n) algorithm described in
059 * William R. Knight's 1966 paper "A Computer Method for Calculating
060 * Kendall's Tau with Ungrouped Data" in the Journal of the American
061 * Statistical Association.
062 *
063 * @see <a href="http://en.wikipedia.org/wiki/Kendall_tau_rank_correlation_coefficient">
064 * Kendall tau rank correlation coefficient (Wikipedia)</a>
065 * @see <a href="http://www.jstor.org/stable/2282833">A Computer
066 * Method for Calculating Kendall's Tau with Ungrouped Data</a>
067 *
068 * @version $Id: KendallsCorrelation.html 885258 2013-11-03 02:46:49Z tn $
069 * @since 3.3
070 */
071public class KendallsCorrelation {
072
073    /** correlation matrix */
074    private final RealMatrix correlationMatrix;
075
076    /**
077     * Create a KendallsCorrelation instance without data.
078     */
079    public KendallsCorrelation() {
080        correlationMatrix = null;
081    }
082
083    /**
084     * Create a KendallsCorrelation from a rectangular array
085     * whose columns represent values of variables to be correlated.
086     *
087     * @param data rectangular array with columns representing variables
088     * @throws IllegalArgumentException if the input data array is not
089     * rectangular with at least two rows and two columns.
090     */
091    public KendallsCorrelation(double[][] data) {
092        this(MatrixUtils.createRealMatrix(data));
093    }
094
095    /**
096     * Create a KendallsCorrelation from a RealMatrix whose columns
097     * represent variables to be correlated.
098     *
099     * @param matrix matrix with columns representing variables to correlate
100     */
101    public KendallsCorrelation(RealMatrix matrix) {
102        correlationMatrix = computeCorrelationMatrix(matrix);
103    }
104
105    /**
106     * Returns the correlation matrix.
107     *
108     * @return correlation matrix
109     */
110    public RealMatrix getCorrelationMatrix() {
111        return correlationMatrix;
112    }
113
114    /**
115     * Computes the Kendall's Tau rank correlation matrix for the columns of
116     * the input matrix.
117     *
118     * @param matrix matrix with columns representing variables to correlate
119     * @return correlation matrix
120     */
121    public RealMatrix computeCorrelationMatrix(final RealMatrix matrix) {
122        int nVars = matrix.getColumnDimension();
123        RealMatrix outMatrix = new BlockRealMatrix(nVars, nVars);
124        for (int i = 0; i < nVars; i++) {
125            for (int j = 0; j < i; j++) {
126                double corr = correlation(matrix.getColumn(i), matrix.getColumn(j));
127                outMatrix.setEntry(i, j, corr);
128                outMatrix.setEntry(j, i, corr);
129            }
130            outMatrix.setEntry(i, i, 1d);
131        }
132        return outMatrix;
133    }
134
135    /**
136     * Computes the Kendall's Tau rank correlation matrix for the columns of
137     * the input rectangular array.  The columns of the array represent values
138     * of variables to be correlated.
139     *
140     * @param matrix matrix with columns representing variables to correlate
141     * @return correlation matrix
142     */
143    public RealMatrix computeCorrelationMatrix(final double[][] matrix) {
144       return computeCorrelationMatrix(new BlockRealMatrix(matrix));
145    }
146
147    /**
148     * Computes the Kendall's Tau rank correlation coefficient between the two arrays.
149     *
150     * @param xArray first data array
151     * @param yArray second data array
152     * @return Returns Kendall's Tau rank correlation coefficient for the two arrays
153     * @throws DimensionMismatchException if the arrays lengths do not match
154     */
155    public double correlation(final double[] xArray, final double[] yArray)
156            throws DimensionMismatchException {
157
158        if (xArray.length != yArray.length) {
159            throw new DimensionMismatchException(xArray.length, yArray.length);
160        }
161
162        final int n = xArray.length;
163        final int numPairs = n * (n - 1) / 2;
164
165        @SuppressWarnings("unchecked")
166        Pair<Double, Double>[] pairs = new Pair[n];
167        for (int i = 0; i < n; i++) {
168            pairs[i] = new Pair<Double, Double>(xArray[i], yArray[i]);
169        }
170
171        Arrays.sort(pairs, new Comparator<Pair<Double, Double>>() {
172            @Override
173            public int compare(Pair<Double, Double> pair1, Pair<Double, Double> pair2) {
174                int compareFirst = pair1.getFirst().compareTo(pair2.getFirst());
175                return compareFirst != 0 ? compareFirst : pair1.getSecond().compareTo(pair2.getSecond());
176            }
177        });
178
179        int tiedXPairs = 0;
180        int tiedXYPairs = 0;
181        int consecutiveXTies = 1;
182        int consecutiveXYTies = 1;
183        Pair<Double, Double> prev = pairs[0];
184        for (int i = 1; i < n; i++) {
185            final Pair<Double, Double> curr = pairs[i];
186            if (curr.getFirst().equals(prev.getFirst())) {
187                consecutiveXTies++;
188                if (curr.getSecond().equals(prev.getSecond())) {
189                    consecutiveXYTies++;
190                } else {
191                    tiedXYPairs += consecutiveXYTies * (consecutiveXYTies - 1) / 2;
192                    consecutiveXYTies = 1;
193                }
194            } else {
195                tiedXPairs += consecutiveXTies * (consecutiveXTies - 1) / 2;
196                consecutiveXTies = 1;
197                tiedXYPairs += consecutiveXYTies * (consecutiveXYTies - 1) / 2;
198                consecutiveXYTies = 1;
199            }
200            prev = curr;
201        }
202        tiedXPairs += consecutiveXTies * (consecutiveXTies - 1) / 2;
203        tiedXYPairs += consecutiveXYTies * (consecutiveXYTies - 1) / 2;
204
205        int swaps = 0;
206        @SuppressWarnings("unchecked")
207        Pair<Double, Double>[] pairsDestination = new Pair[n];
208        for (int segmentSize = 1; segmentSize < n; segmentSize <<= 1) {
209            for (int offset = 0; offset < n; offset += 2 * segmentSize) {
210                int i = offset;
211                final int iEnd = FastMath.min(i + segmentSize, n);
212                int j = iEnd;
213                final int jEnd = FastMath.min(j + segmentSize, n);
214
215                int copyLocation = offset;
216                while (i < iEnd || j < jEnd) {
217                    if (i < iEnd) {
218                        if (j < jEnd) {
219                            if (pairs[i].getSecond().compareTo(pairs[j].getSecond()) <= 0) {
220                                pairsDestination[copyLocation] = pairs[i];
221                                i++;
222                            } else {
223                                pairsDestination[copyLocation] = pairs[j];
224                                j++;
225                                swaps += iEnd - i;
226                            }
227                        } else {
228                            pairsDestination[copyLocation] = pairs[i];
229                            i++;
230                        }
231                    } else {
232                        pairsDestination[copyLocation] = pairs[j];
233                        j++;
234                    }
235                    copyLocation++;
236                }
237            }
238            final Pair<Double, Double>[] pairsTemp = pairs;
239            pairs = pairsDestination;
240            pairsDestination = pairsTemp;
241        }
242
243        int tiedYPairs = 0;
244        int consecutiveYTies = 1;
245        prev = pairs[0];
246        for (int i = 1; i < n; i++) {
247            final Pair<Double, Double> curr = pairs[i];
248            if (curr.getSecond().equals(prev.getSecond())) {
249                consecutiveYTies++;
250            } else {
251                tiedYPairs += consecutiveYTies * (consecutiveYTies - 1) / 2;
252                consecutiveYTies = 1;
253            }
254            prev = curr;
255        }
256        tiedYPairs += consecutiveYTies * (consecutiveYTies - 1) / 2;
257
258        int concordantMinusDiscordant = numPairs - tiedXPairs - tiedYPairs + tiedXYPairs - 2 * swaps;
259        return concordantMinusDiscordant / FastMath.sqrt((numPairs - tiedXPairs) * (numPairs - tiedYPairs));
260    }
261}