1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2018 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
4 | *
|
---|
5 | * This file is part of HeuristicLab.
|
---|
6 | *
|
---|
7 | * HeuristicLab is free software: you can redistribute it and/or modify
|
---|
8 | * it under the terms of the GNU General Public License as published by
|
---|
9 | * the Free Software Foundation, either version 3 of the License, or
|
---|
10 | * (at your option) any later version.
|
---|
11 | *
|
---|
12 | * HeuristicLab is distributed in the hope that it will be useful,
|
---|
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
15 | * GNU General Public License for more details.
|
---|
16 | *
|
---|
17 | * You should have received a copy of the GNU General Public License
|
---|
18 | * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
|
---|
19 | */
|
---|
20 | #endregion
|
---|
21 | using System;
|
---|
22 | using System.Collections.Generic;
|
---|
23 | using System.Linq;
|
---|
24 | using HeuristicLab.Common;
|
---|
25 |
|
---|
26 | namespace HeuristicLab.Optimization {
|
---|
27 | public static class HypervolumeCalculator {
|
---|
28 | public static double[] CalculateNadirPoint(IEnumerable<double[]> qualities, bool[] maximization) {
|
---|
29 | var res = maximization.Select(m => m ? double.MaxValue : double.MinValue).ToArray();
|
---|
30 | foreach (var quality in qualities)
|
---|
31 | for (var i = 0; i < quality.Length; i++)
|
---|
32 | if (maximization[i] == res[i] > quality[i])
|
---|
33 | res[i] = quality[i];
|
---|
34 | return res;
|
---|
35 | }
|
---|
36 |
|
---|
37 | /// <summary>
|
---|
38 | /// The Hypervolume-metric is defined as the HypervolumeCalculator enclosed between a given reference point,
|
---|
39 | /// that is fixed for every evaluation function and the evaluated qualities.
|
---|
40 | ///
|
---|
41 | /// Example:
|
---|
42 | /// r is the reference point at (1|1) and every point p is part of the evaluated qualities
|
---|
43 | /// The filled area labled HV is the 2 dimensional HypervolumeCalculator enclosed by this qualities.
|
---|
44 | ///
|
---|
45 | /// (0|1) (1|1)
|
---|
46 | /// + +-------------r
|
---|
47 | /// | |###### HV ###|
|
---|
48 | /// | p------+######|
|
---|
49 | /// | p+#####|
|
---|
50 | /// | |#####|
|
---|
51 | /// | p-+###|
|
---|
52 | /// | p---+
|
---|
53 | /// |
|
---|
54 | /// +--------------------1
|
---|
55 | /// (0|0) (1|0)
|
---|
56 | ///
|
---|
57 | /// Please note that in this example both dimensions are minimized. The reference point needs to be dominated by EVERY point in the evaluated qualities
|
---|
58 | ///
|
---|
59 | /// </summary>
|
---|
60 | ///
|
---|
61 | public static double CalculateHypervolume(IList<double[]> qualities, double[] referencePoint, bool[] maximization) {
|
---|
62 | qualities = qualities.Where(vec => DominationCalculator.Dominates(vec, referencePoint, maximization, false) == DominationResult.Dominates).ToArray();
|
---|
63 | if (qualities.Count== 0) return 0; //TODO computation for negative hypervolume?
|
---|
64 | if (maximization.Length == 2)
|
---|
65 | return Calculate2D(qualities, referencePoint, maximization);
|
---|
66 |
|
---|
67 | if (Array.TrueForAll(maximization, x => !x))
|
---|
68 | return CalculateMultiDimensional(qualities, referencePoint);
|
---|
69 | throw new NotImplementedException("HypervolumeCalculator calculation for more than two dimensions is supported only with minimization problems.");
|
---|
70 | }
|
---|
71 |
|
---|
72 |
|
---|
73 | /// <summary>
|
---|
74 | /// Caluclates the Hypervolume for a 2 dimensional problem
|
---|
75 | /// </summary>
|
---|
76 | /// <param name="front">All points within the front need to be Non-Dominated and need to dominate the reference point</param>
|
---|
77 | /// <param name="referencePoint"></param>
|
---|
78 | /// <param name="maximization"></param>
|
---|
79 | /// <returns></returns>
|
---|
80 | public static double Calculate2D(IList<double[]> front, double[] referencePoint, bool[] maximization) {
|
---|
81 | if (front == null) throw new ArgumentNullException("front");
|
---|
82 | if (referencePoint == null) throw new ArgumentNullException("referencePoint");
|
---|
83 | if (maximization == null) throw new ArgumentNullException("maximization");
|
---|
84 | if (!front.Any()) throw new ArgumentException("Front must not be empty.");
|
---|
85 | if (referencePoint.Length != 2) throw new ArgumentException("ReferencePoint must have exactly two dimensions.");
|
---|
86 |
|
---|
87 | var set = front.ToArray();
|
---|
88 | if (set.Any(s => s.Length != 2)) throw new ArgumentException("Points in qualities must have exactly two dimensions.");
|
---|
89 |
|
---|
90 | Array.Sort(set, new DimensionComparer(0, maximization[0]));
|
---|
91 |
|
---|
92 | double sum = 0;
|
---|
93 | for (var i = 0; i < set.Length - 1; i++)
|
---|
94 | sum += Math.Abs(set[i][0] - set[i + 1][0]) * Math.Abs(set[i][1] - referencePoint[1]);
|
---|
95 | var lastPoint = set[set.Length - 1];
|
---|
96 | sum += Math.Abs(lastPoint[0] - referencePoint[0]) * Math.Abs(lastPoint[1] - referencePoint[1]);
|
---|
97 |
|
---|
98 | return sum;
|
---|
99 | }
|
---|
100 |
|
---|
101 | public static double CalculateMultiDimensional(IList<double[]> front, double[] referencePoint) {
|
---|
102 | if (referencePoint == null || referencePoint.Length < 3) throw new ArgumentException("ReferencePoint unfit for complex HypervolumeCalculator calculation");
|
---|
103 |
|
---|
104 | var objectives = referencePoint.Length;
|
---|
105 | var fronList = front.ToList();
|
---|
106 | fronList.StableSort(new DimensionComparer(objectives - 1, false));
|
---|
107 |
|
---|
108 | var regLow = Enumerable.Repeat(1E15, objectives).ToArray();
|
---|
109 | foreach (var p in fronList) {
|
---|
110 | for (var i = 0; i < regLow.Length; i++) {
|
---|
111 | if (p[i] < regLow[i]) regLow[i] = p[i];
|
---|
112 | }
|
---|
113 | }
|
---|
114 |
|
---|
115 | return Stream(regLow, referencePoint, fronList, 0, referencePoint[objectives - 1], (int)Math.Sqrt(fronList.Count), objectives);
|
---|
116 | }
|
---|
117 |
|
---|
118 |
|
---|
119 | //within Stream a number of equality comparisons on double values are performed
|
---|
120 | //this is intentional and required
|
---|
121 | private static double Stream(double[] regionLow, double[] regionUp, List<double[]> front, int split, double cover, int sqrtNoPoints, int objectives) {
|
---|
122 | var coverOld = cover;
|
---|
123 | var coverIndex = 0;
|
---|
124 | var coverIndexOld = -1;
|
---|
125 | int c;
|
---|
126 | double result = 0;
|
---|
127 |
|
---|
128 | var dMeasure = GetMeasure(regionLow, regionUp, objectives);
|
---|
129 | while (cover == coverOld && coverIndex < front.Count()) {
|
---|
130 | if (coverIndexOld == coverIndex) break;
|
---|
131 | coverIndexOld = coverIndex;
|
---|
132 | if (Covers(front[coverIndex], regionLow, objectives)) {
|
---|
133 | cover = front[coverIndex][objectives - 1];
|
---|
134 | result += dMeasure * (coverOld - cover);
|
---|
135 | } else coverIndex++;
|
---|
136 | }
|
---|
137 |
|
---|
138 | for (c = coverIndex; c > 0; c--) if (front[c - 1][objectives - 1] == cover) coverIndex--;
|
---|
139 | if (coverIndex == 0) return result;
|
---|
140 |
|
---|
141 | var allPiles = true;
|
---|
142 | var piles = new int[coverIndex];
|
---|
143 | for (var i = 0; i < coverIndex; i++) {
|
---|
144 | piles[i] = IsPile(front[i], regionLow, objectives);
|
---|
145 | if (piles[i] == -1) {
|
---|
146 | allPiles = false;
|
---|
147 | break;
|
---|
148 | }
|
---|
149 | }
|
---|
150 |
|
---|
151 | if (allPiles) {
|
---|
152 | var trellis = new double[regionUp.Length];
|
---|
153 | for (var j = 0; j < trellis.Length; j++) trellis[j] = regionUp[j];
|
---|
154 | double next;
|
---|
155 | var i = 0;
|
---|
156 | do {
|
---|
157 | var current = front[i][objectives - 1];
|
---|
158 | do {
|
---|
159 | if (front[i][piles[i]] < trellis[piles[i]]) trellis[piles[i]] = front[i][piles[i]];
|
---|
160 | i++;
|
---|
161 | if (i < coverIndex) next = front[i][objectives - 1];
|
---|
162 | else {
|
---|
163 | next = cover;
|
---|
164 | break;
|
---|
165 | }
|
---|
166 | }
|
---|
167 | while (next == current);
|
---|
168 | result += ComputeTrellis(regionLow, regionUp, trellis, objectives) * (next - current);
|
---|
169 | }
|
---|
170 | while (next != cover);
|
---|
171 | } else {
|
---|
172 | double bound = -1;
|
---|
173 | var boundaries = new double[coverIndex];
|
---|
174 | var noBoundaries = new double[coverIndex];
|
---|
175 | var boundIdx = 0;
|
---|
176 | var noBoundIdx = 0;
|
---|
177 |
|
---|
178 | do {
|
---|
179 | for (var i = 0; i < coverIndex; i++) {
|
---|
180 | var contained = ContainesBoundary(front[i], regionLow, split);
|
---|
181 | if (contained == 0) boundaries[boundIdx++] = front[i][split];
|
---|
182 | else if (contained == 1) noBoundaries[noBoundIdx++] = front[i][split];
|
---|
183 | }
|
---|
184 | if (boundIdx > 0) bound = GetMedian(boundaries, boundIdx);
|
---|
185 | else if (noBoundIdx > sqrtNoPoints) bound = GetMedian(noBoundaries, noBoundIdx);
|
---|
186 | else split++;
|
---|
187 | }
|
---|
188 | while (bound == -1.0);
|
---|
189 |
|
---|
190 | var pointsChildLow = new List<double[]>();
|
---|
191 | var pointsChildUp = new List<double[]>();
|
---|
192 | var regionUpC = new double[regionUp.Length];
|
---|
193 | for (var j = 0; j < regionUpC.Length; j++) regionUpC[j] = regionUp[j];
|
---|
194 | var regionLowC = new double[regionLow.Length];
|
---|
195 | for (var j = 0; j < regionLowC.Length; j++) regionLowC[j] = regionLow[j];
|
---|
196 |
|
---|
197 | for (var i = 0; i < coverIndex; i++) {
|
---|
198 | if (PartCovers(front[i], regionUpC, objectives)) pointsChildUp.Add(front[i]);
|
---|
199 | if (PartCovers(front[i], regionUp, objectives)) pointsChildLow.Add(front[i]);
|
---|
200 | }
|
---|
201 |
|
---|
202 | if (pointsChildUp.Count > 0) result += Stream(regionLow, regionUpC, pointsChildUp, split, cover, sqrtNoPoints, objectives);
|
---|
203 | if (pointsChildLow.Count > 0) result += Stream(regionLowC, regionUp, pointsChildLow, split, cover, sqrtNoPoints, objectives);
|
---|
204 | }
|
---|
205 | return result;
|
---|
206 | }
|
---|
207 |
|
---|
208 | private static double GetMedian(double[] vector, int length) {
|
---|
209 | return vector.Take(length).Median();
|
---|
210 | }
|
---|
211 |
|
---|
212 | private static double ComputeTrellis(double[] regionLow, double[] regionUp, double[] trellis, int objectives) {
|
---|
213 | var bs = new bool[objectives - 1];
|
---|
214 | for (var i = 0; i < bs.Length; i++) bs[i] = true;
|
---|
215 |
|
---|
216 | double result = 0;
|
---|
217 | var noSummands = BinarayToInt(bs);
|
---|
218 | for (uint i = 1; i <= noSummands; i++) {
|
---|
219 | double summand = 1;
|
---|
220 | IntToBinary(i, bs);
|
---|
221 | var oneCounter = 0;
|
---|
222 | for (var j = 0; j < objectives - 1; j++) {
|
---|
223 | if (bs[j]) {
|
---|
224 | summand *= regionUp[j] - trellis[j];
|
---|
225 | oneCounter++;
|
---|
226 | } else {
|
---|
227 | summand *= regionUp[j] - regionLow[j];
|
---|
228 | }
|
---|
229 | }
|
---|
230 | if (oneCounter % 2 == 0) result -= summand;
|
---|
231 | else result += summand;
|
---|
232 | }
|
---|
233 | return result;
|
---|
234 | }
|
---|
235 |
|
---|
236 | private static void IntToBinary(uint i, bool[] bs) {
|
---|
237 | for (var j = 0; j < bs.Length; j++) bs[j] = false;
|
---|
238 | var rest = i;
|
---|
239 | var idx = 0;
|
---|
240 | while (rest != 0) {
|
---|
241 | bs[idx] = rest % 2 == 1;
|
---|
242 | rest = rest / 2;
|
---|
243 | idx++;
|
---|
244 | }
|
---|
245 | }
|
---|
246 |
|
---|
247 | private static uint BinarayToInt(bool[] bs) {
|
---|
248 | uint result = 0;
|
---|
249 | for (var i = 0; i < bs.Length; i++) {
|
---|
250 | result += bs[i] ? ((uint)1 << i) : 0;
|
---|
251 | }
|
---|
252 | return result;
|
---|
253 | }
|
---|
254 |
|
---|
255 | private static int IsPile(double[] cuboid, double[] regionLow, int objectives) {
|
---|
256 | var pile = cuboid.Length;
|
---|
257 | for (var i = 0; i < objectives - 1; i++) {
|
---|
258 | if (cuboid[i] > regionLow[i]) {
|
---|
259 | if (pile != objectives) return 1;
|
---|
260 | pile = i;
|
---|
261 | }
|
---|
262 | }
|
---|
263 | return pile;
|
---|
264 | }
|
---|
265 |
|
---|
266 | private static double GetMeasure(double[] regionLow, double[] regionUp, int objectives) {
|
---|
267 | double volume = 1;
|
---|
268 | for (var i = 0; i < objectives - 1; i++) {
|
---|
269 | volume *= (regionUp[i] - regionLow[i]);
|
---|
270 | }
|
---|
271 | return volume;
|
---|
272 | }
|
---|
273 |
|
---|
274 | private static int ContainesBoundary(double[] cub, double[] regionLow, int split) {
|
---|
275 | if (regionLow[split] >= cub[split]) return -1;
|
---|
276 | else {
|
---|
277 | for (var j = 0; j < split; j++) {
|
---|
278 | if (regionLow[j] < cub[j]) return 1;
|
---|
279 | }
|
---|
280 | }
|
---|
281 | return 0;
|
---|
282 | }
|
---|
283 |
|
---|
284 | private static bool PartCovers(double[] v, double[] regionUp, int objectives) {
|
---|
285 | for (var i = 0; i < objectives - 1; i++) {
|
---|
286 | if (v[i] >= regionUp[i]) return false;
|
---|
287 | }
|
---|
288 | return true;
|
---|
289 | }
|
---|
290 |
|
---|
291 | private static bool Covers(double[] v, double[] regionLow, int objectives) {
|
---|
292 | for (var i = 0; i < objectives - 1; i++) {
|
---|
293 | if (v[i] > regionLow[i]) return false;
|
---|
294 | }
|
---|
295 | return true;
|
---|
296 | }
|
---|
297 |
|
---|
298 | private class DimensionComparer : IComparer<double[]> {
|
---|
299 | private readonly int dimension;
|
---|
300 | private readonly int descending;
|
---|
301 |
|
---|
302 | public DimensionComparer(int dimension, bool descending) {
|
---|
303 | this.dimension = dimension;
|
---|
304 | this.descending = descending ? -1 : 1;
|
---|
305 | }
|
---|
306 |
|
---|
307 | public int Compare(double[] x, double[] y) {
|
---|
308 | return x[dimension].CompareTo(y[dimension]) * descending;
|
---|
309 | }
|
---|
310 | }
|
---|
311 | }
|
---|
312 | } |
---|