1 | package negotiator.boaframework.offeringstrategy.anac2012;
|
---|
2 |
|
---|
3 | import java.util.ArrayList;
|
---|
4 | import java.util.Map;
|
---|
5 | import java.util.Random;
|
---|
6 |
|
---|
7 | import agents.Jama.Matrix;
|
---|
8 | import agents.anac.y2012.IAMhaggler2012.agents2011.southampton.utils.RandomBidCreator;
|
---|
9 | import agents.anac.y2012.IAMhaggler2012.utility.SouthamptonUtilitySpace;
|
---|
10 | import agents.org.apache.commons.math.MathException;
|
---|
11 | import agents.org.apache.commons.math.MaxIterationsExceededException;
|
---|
12 | import agents.org.apache.commons.math.special.Erf;
|
---|
13 | import agents.uk.ac.soton.ecs.gp4j.bmc.BasicPrior;
|
---|
14 | import agents.uk.ac.soton.ecs.gp4j.bmc.GaussianProcessMixture;
|
---|
15 | import agents.uk.ac.soton.ecs.gp4j.bmc.GaussianProcessMixturePrediction;
|
---|
16 | import agents.uk.ac.soton.ecs.gp4j.bmc.GaussianProcessRegressionBMC;
|
---|
17 | import agents.uk.ac.soton.ecs.gp4j.gp.covariancefunctions.CovarianceFunction;
|
---|
18 | import agents.uk.ac.soton.ecs.gp4j.gp.covariancefunctions.Matern3CovarianceFunction;
|
---|
19 | import agents.uk.ac.soton.ecs.gp4j.gp.covariancefunctions.NoiseCovarianceFunction;
|
---|
20 | import agents.uk.ac.soton.ecs.gp4j.gp.covariancefunctions.SumCovarianceFunction;
|
---|
21 | import genius.core.Bid;
|
---|
22 | import genius.core.BidIterator;
|
---|
23 | import genius.core.bidding.BidDetails;
|
---|
24 | import genius.core.boaframework.NegotiationSession;
|
---|
25 | import genius.core.boaframework.NoModel;
|
---|
26 | import genius.core.boaframework.OMStrategy;
|
---|
27 | import genius.core.boaframework.OfferingStrategy;
|
---|
28 | import genius.core.boaframework.OpponentModel;
|
---|
29 | import genius.core.boaframework.SortedOutcomeSpace;
|
---|
30 | import genius.core.misc.Pair;
|
---|
31 | import genius.core.utility.AdditiveUtilitySpace;
|
---|
32 | import negotiator.boaframework.opponentmodel.DefaultModel;
|
---|
33 |
|
---|
34 | /**
|
---|
35 | * @author Colin Williams
|
---|
36 | *
|
---|
37 | * The IAMhaggler Agent, created for ANAC 2012. Designed by C. R.
|
---|
38 | * Williams, V. Robu, E. H. Gerding and N. R. Jennings.
|
---|
39 | *
|
---|
40 | */
|
---|
41 | public class IAMHaggler2012_Offering extends OfferingStrategy {
|
---|
42 |
|
---|
43 | protected double RISK_PARAMETER = 1;
|
---|
44 |
|
---|
45 | private Matrix utilitySamples;
|
---|
46 | private Matrix timeSamples;
|
---|
47 | private Matrix utility;
|
---|
48 | private GaussianProcessRegressionBMC regression;
|
---|
49 | private double lastRegressionTime = 0;
|
---|
50 | private double lastRegressionUtility = 1;
|
---|
51 | private ArrayList<Double> opponentTimes = new ArrayList<Double>();
|
---|
52 | private ArrayList<Double> opponentUtilities = new ArrayList<Double>();
|
---|
53 |
|
---|
54 | private double maxUtilityInTimeSlot;
|
---|
55 | private int lastTimeSlot = -1;
|
---|
56 | private Matrix means;
|
---|
57 | private Matrix variances;
|
---|
58 |
|
---|
59 | private double maxUtility;
|
---|
60 | private Bid bestReceivedBid;
|
---|
61 | private double previousTargetUtility;
|
---|
62 | private double intercept;
|
---|
63 | private Matrix matrixTimeSamplesAdjust;
|
---|
64 | private double maxOfferedUtility = Double.MIN_VALUE;
|
---|
65 | private double minOfferedUtility = Double.MAX_VALUE;
|
---|
66 | private SortedOutcomeSpace outcomespace;
|
---|
67 | private RandomBidCreator bidCreator;
|
---|
68 | private AdditiveUtilitySpace utilitySpace;
|
---|
69 | private SouthamptonUtilitySpace sus;
|
---|
70 |
|
---|
71 | /**
|
---|
72 | * Empty constructor for the BOA framework.
|
---|
73 | */
|
---|
74 | public IAMHaggler2012_Offering() {
|
---|
75 | }
|
---|
76 |
|
---|
77 | public IAMHaggler2012_Offering(NegotiationSession negoSession, OpponentModel model, OMStrategy oms)
|
---|
78 | throws Exception {
|
---|
79 | init(negoSession, model, oms, null);
|
---|
80 | }
|
---|
81 |
|
---|
82 | /**
|
---|
83 | * Init required for the Decoupled Framework.
|
---|
84 | */
|
---|
85 | @Override
|
---|
86 | public void init(NegotiationSession negoSession, OpponentModel model, OMStrategy oms,
|
---|
87 | Map<String, Double> parameters) throws Exception {
|
---|
88 | if (model instanceof DefaultModel) {
|
---|
89 | model = new NoModel();
|
---|
90 | }
|
---|
91 | super.init(negoSession, model, oms, parameters);
|
---|
92 | if (!(opponentModel instanceof NoModel)) {
|
---|
93 | outcomespace = new SortedOutcomeSpace(negotiationSession.getUtilitySpace());
|
---|
94 | }
|
---|
95 |
|
---|
96 | utilitySpace = (AdditiveUtilitySpace) negoSession.getUtilitySpace();
|
---|
97 | double discountingFactor = 0.5;
|
---|
98 | try {
|
---|
99 | discountingFactor = adjustDiscountFactor(utilitySpace.getDiscountFactor());
|
---|
100 | } catch (Exception ex) {
|
---|
101 | ex.printStackTrace();
|
---|
102 | }
|
---|
103 | if (discountingFactor == 0)
|
---|
104 | discountingFactor = 1;
|
---|
105 | makeUtilitySamples(100);
|
---|
106 | makeTimeSamples(100);
|
---|
107 | Matrix discounting = generateDiscountingFunction(discountingFactor);
|
---|
108 | Matrix risk = generateRiskFunction(RISK_PARAMETER);
|
---|
109 | utility = risk.arrayTimes(discounting);
|
---|
110 |
|
---|
111 | BasicPrior[] bps = { new BasicPrior(11, 0.252, 0.5), new BasicPrior(11, 0.166, 0.5),
|
---|
112 | new BasicPrior(1, .01, 1.0) };
|
---|
113 | CovarianceFunction cf = new SumCovarianceFunction(Matern3CovarianceFunction.getInstance(),
|
---|
114 | NoiseCovarianceFunction.getInstance());
|
---|
115 |
|
---|
116 | regression = new GaussianProcessRegressionBMC();
|
---|
117 | regression.setCovarianceFunction(cf);
|
---|
118 | regression.setPriors(bps);
|
---|
119 |
|
---|
120 | // regression.calculateRegression(new Matrix(new double[] {}, 0), new
|
---|
121 | // Matrix(new double[] {}, 0));
|
---|
122 |
|
---|
123 | maxUtility = 0;
|
---|
124 | previousTargetUtility = 1;
|
---|
125 | bidCreator = new RandomBidCreator();
|
---|
126 |
|
---|
127 | sus = new SouthamptonUtilitySpace(utilitySpace);
|
---|
128 | }
|
---|
129 |
|
---|
130 | @Override
|
---|
131 | public BidDetails determineOpeningBid() {
|
---|
132 | return determineNextBid();
|
---|
133 | }
|
---|
134 |
|
---|
135 | @Override
|
---|
136 | public BidDetails determineNextBid() {
|
---|
137 | Bid opponentBid = negotiationSession.getOpponentBidHistory().getLastBid();
|
---|
138 | Bid b;
|
---|
139 | try {
|
---|
140 | b = handleOffer(opponentBid);
|
---|
141 | nextBid = new BidDetails(b, negotiationSession.getUtilitySpace().getUtility(b));
|
---|
142 | } catch (Exception e) {
|
---|
143 | e.printStackTrace();
|
---|
144 | }
|
---|
145 |
|
---|
146 | return nextBid;
|
---|
147 | }
|
---|
148 |
|
---|
149 | /**
|
---|
150 | * Handle an opponent's offer.
|
---|
151 | *
|
---|
152 | * @param opponentBid
|
---|
153 | * The bid made by the opponent.
|
---|
154 | * @return the action that we should take in response to the opponent's
|
---|
155 | * offer.
|
---|
156 | * @throws Exception
|
---|
157 | */
|
---|
158 | private Bid handleOffer(Bid opponentBid) throws Exception {
|
---|
159 | Bid bidToOffer;
|
---|
160 | if (negotiationSession.getOwnBidHistory().getHistory().isEmpty()) {
|
---|
161 | // Special case to handle first action
|
---|
162 | bidToOffer = proposeInitialBid();
|
---|
163 | if (bidToOffer == null) {
|
---|
164 | endNegotiation = true;
|
---|
165 | }
|
---|
166 | } else {
|
---|
167 | bidToOffer = proposeNextBid(opponentBid);
|
---|
168 | if (bidToOffer == null)
|
---|
169 | endNegotiation = true;
|
---|
170 | }
|
---|
171 |
|
---|
172 | return bidToOffer;
|
---|
173 | }
|
---|
174 |
|
---|
175 | /*
|
---|
176 | *
|
---|
177 | * (non-Javadoc)
|
---|
178 | *
|
---|
179 | * @see agents2011.southampton.IAMhaggler2011#proposeInitialBid()
|
---|
180 | */
|
---|
181 | private Bid proposeInitialBid() throws Exception {
|
---|
182 | Bid b = sus.getMaxUtilityBid();
|
---|
183 | if (utilitySpace.getUtilityWithDiscount(b, negotiationSession.getTimeline()) < utilitySpace
|
---|
184 | .getReservationValueWithDiscount(negotiationSession.getTimeline())) {
|
---|
185 | return null;
|
---|
186 | }
|
---|
187 | return b;
|
---|
188 | }
|
---|
189 |
|
---|
190 | /*
|
---|
191 | * (non-Javadoc)
|
---|
192 | *
|
---|
193 | * @see agents.southampton.SouthamptonAgent#proposeNextBid(negotiator.Bid)
|
---|
194 | */
|
---|
195 | protected Bid proposeNextBid(Bid opponentBid) throws Exception {
|
---|
196 |
|
---|
197 | double opponentUtility = utilitySpace.getUtility(opponentBid);
|
---|
198 |
|
---|
199 | if (opponentUtility > maxUtility) {
|
---|
200 | bestReceivedBid = opponentBid;
|
---|
201 | maxUtility = opponentUtility;
|
---|
202 | }
|
---|
203 |
|
---|
204 | double targetUtility = getTarget(opponentUtility, negotiationSession.getTime());
|
---|
205 |
|
---|
206 | if (targetUtility <= maxUtility && previousTargetUtility > maxUtility)
|
---|
207 | return bestReceivedBid;
|
---|
208 | previousTargetUtility = targetUtility;
|
---|
209 |
|
---|
210 | Bid b = null;
|
---|
211 | if (opponentModel instanceof NoModel) {
|
---|
212 | // Now get a random bid in the range targetUtility � 0.025
|
---|
213 | b = bidCreator.getBid(utilitySpace, targetUtility - 0.025, targetUtility + 0.025);
|
---|
214 | } else {
|
---|
215 | b = omStrategy.getBid(outcomespace, targetUtility).getBid();
|
---|
216 | }
|
---|
217 |
|
---|
218 | if (utilitySpace.getUtilityWithDiscount(b, negotiationSession.getTimeline()) < utilitySpace
|
---|
219 | .getReservationValueWithDiscount(negotiationSession.getTimeline())) {
|
---|
220 | return utilitySpace.getMaxUtilityBid();
|
---|
221 | }
|
---|
222 | // System.out.println("Decoupled Bid: " + b);
|
---|
223 |
|
---|
224 | return b;
|
---|
225 | }
|
---|
226 |
|
---|
227 | /*
|
---|
228 | * (non-Javadoc)
|
---|
229 | *
|
---|
230 | * @see agents2011.southampton.IAMhaggler2011#getTarget(double, double)
|
---|
231 | */
|
---|
232 | protected double getTarget(double opponentUtility, double time) {
|
---|
233 |
|
---|
234 | maxOfferedUtility = Math.max(maxOfferedUtility, opponentUtility);
|
---|
235 | minOfferedUtility = Math.min(minOfferedUtility, opponentUtility);
|
---|
236 |
|
---|
237 | // Calculate the current time slot
|
---|
238 | int timeSlot = (int) Math.floor(time * 36);
|
---|
239 |
|
---|
240 | boolean regressionUpdateRequired = false;
|
---|
241 | if (lastTimeSlot == -1) {
|
---|
242 | regressionUpdateRequired = true;
|
---|
243 | }
|
---|
244 |
|
---|
245 | // If the time slot has changed
|
---|
246 | if (timeSlot != lastTimeSlot) {
|
---|
247 | if (lastTimeSlot != -1) {
|
---|
248 | // Store the data from the time slot
|
---|
249 | opponentTimes.add((lastTimeSlot + 0.5) / 36.0);
|
---|
250 | if (opponentUtilities.size() == 0) {
|
---|
251 | intercept = Math.max(0.5, maxUtilityInTimeSlot);
|
---|
252 | double[] timeSamplesAdjust = new double[timeSamples.getColumnDimension()];
|
---|
253 | int i = 0;
|
---|
254 | double gradient = 0.9 - intercept;
|
---|
255 | for (double d : timeSamples.getRowPackedCopy()) {
|
---|
256 | timeSamplesAdjust[i++] = intercept + (gradient * d);
|
---|
257 | }
|
---|
258 | matrixTimeSamplesAdjust = new Matrix(timeSamplesAdjust, timeSamplesAdjust.length);
|
---|
259 | }
|
---|
260 | opponentUtilities.add(maxUtilityInTimeSlot);
|
---|
261 | // Flag regression receiveMessage required
|
---|
262 | regressionUpdateRequired = true;
|
---|
263 | }
|
---|
264 | // Update the time slot
|
---|
265 | lastTimeSlot = timeSlot;
|
---|
266 | // Reset the max utility
|
---|
267 | maxUtilityInTimeSlot = 0;
|
---|
268 | }
|
---|
269 |
|
---|
270 | // Calculate the maximum utility observed in the current time slot
|
---|
271 | maxUtilityInTimeSlot = Math.max(maxUtilityInTimeSlot, opponentUtility);
|
---|
272 |
|
---|
273 | if (timeSlot == 0) {
|
---|
274 | return 1.0 - time / 2.0;
|
---|
275 | }
|
---|
276 |
|
---|
277 | if (regressionUpdateRequired) {
|
---|
278 | double gradient = 0.9 - intercept;
|
---|
279 | /*
|
---|
280 | * double[] x = new double[opponentTimes.size()]; double[] yAdjust =
|
---|
281 | * new double[opponentTimes.size()]; double[] y = new
|
---|
282 | * double[opponentUtilities.size()];
|
---|
283 | *
|
---|
284 | * int i; i = 0; for (double d : opponentTimes) { x[i++] = d; } i =
|
---|
285 | * 0; for (double d : opponentTimes) { yAdjust[i++] = intercept +
|
---|
286 | * (gradient * d); } i = 0; for (double d : opponentUtilities) {
|
---|
287 | * y[i++] = d; }
|
---|
288 | *
|
---|
289 | * Matrix matrixX = new Matrix(x, x.length); Matrix matrixYAdjust =
|
---|
290 | * new Matrix(yAdjust, yAdjust.length); Matrix matrixY = new
|
---|
291 | * Matrix(y, y.length);
|
---|
292 | *
|
---|
293 | * matrixY.minusEquals(matrixYAdjust);
|
---|
294 | *
|
---|
295 | * //GaussianProcessMixture predictor =
|
---|
296 | * regression.calculateRegression(matrixX, matrixY);
|
---|
297 | */
|
---|
298 |
|
---|
299 | GaussianProcessMixture predictor;
|
---|
300 |
|
---|
301 | if (lastTimeSlot == -1) {
|
---|
302 | predictor = regression.calculateRegression(new double[] {}, new double[] {});
|
---|
303 | } else {
|
---|
304 | double x;
|
---|
305 | double y;
|
---|
306 | try {
|
---|
307 | x = opponentTimes.get(opponentTimes.size() - 1);
|
---|
308 | y = opponentUtilities.get(opponentUtilities.size() - 1);
|
---|
309 | } catch (Exception ex) {
|
---|
310 | System.out.println("Error getting x or y");
|
---|
311 | throw new Error(ex);
|
---|
312 | }
|
---|
313 |
|
---|
314 | predictor = regression.updateRegression(new Matrix(new double[] { x }, 1),
|
---|
315 | new Matrix(new double[] { y - intercept - (gradient * x) }, 1));
|
---|
316 | }
|
---|
317 |
|
---|
318 | GaussianProcessMixturePrediction prediction = predictor.calculatePrediction(timeSamples.transpose());
|
---|
319 |
|
---|
320 | // Store the means and variances
|
---|
321 | means = prediction.getMean().plus(matrixTimeSamplesAdjust);
|
---|
322 | variances = prediction.getVariance();
|
---|
323 |
|
---|
324 | }
|
---|
325 |
|
---|
326 | Pair<Matrix, Matrix> acceptMatrices = generateProbabilityAccept(means, variances, time);
|
---|
327 | Matrix probabilityAccept = acceptMatrices.getFirst();
|
---|
328 | Matrix cumulativeAccept = acceptMatrices.getSecond();
|
---|
329 |
|
---|
330 | Matrix probabilityExpectedUtility = probabilityAccept.arrayTimes(utility);
|
---|
331 | Matrix cumulativeExpectedUtility = cumulativeAccept.arrayTimes(utility);
|
---|
332 |
|
---|
333 | Pair<Double, Double> bestAgreement = getExpectedBestAgreement(probabilityExpectedUtility,
|
---|
334 | cumulativeExpectedUtility, time);
|
---|
335 | double bestTime = bestAgreement.getFirst();
|
---|
336 | double bestUtility = bestAgreement.getSecond();
|
---|
337 |
|
---|
338 | double targetUtility = lastRegressionUtility + ((time - lastRegressionTime)
|
---|
339 | * (bestUtility - lastRegressionUtility) / (bestTime - lastRegressionTime));
|
---|
340 |
|
---|
341 | // Store the target utility and time
|
---|
342 | lastRegressionUtility = targetUtility;
|
---|
343 | lastRegressionTime = time;
|
---|
344 |
|
---|
345 | double d = limitConcession(targetUtility);
|
---|
346 |
|
---|
347 | return Math.max(utilitySpace.getReservationValueWithDiscount(time), d);
|
---|
348 | }
|
---|
349 |
|
---|
350 | /**
|
---|
351 | * Create an m-by-1 matrix of utility samples.
|
---|
352 | *
|
---|
353 | * @param m
|
---|
354 | * The sample size.
|
---|
355 | */
|
---|
356 | private void makeUtilitySamples(int m) {
|
---|
357 | double[] utilitySamplesArray = new double[m];
|
---|
358 | {
|
---|
359 | for (int i = 0; i < utilitySamplesArray.length; i++) {
|
---|
360 | utilitySamplesArray[i] = 1.0 - ((double) i + 0.5) / ((double) m + 1.0);
|
---|
361 | }
|
---|
362 | }
|
---|
363 | utilitySamples = new Matrix(utilitySamplesArray, utilitySamplesArray.length);
|
---|
364 | }
|
---|
365 |
|
---|
366 | /**
|
---|
367 | * Create a 1-by-n matrix of time samples.
|
---|
368 | *
|
---|
369 | * @param n
|
---|
370 | * The sample size.
|
---|
371 | */
|
---|
372 | private void makeTimeSamples(int n) {
|
---|
373 | double[] timeSamplesArray = new double[n + 1];
|
---|
374 | {
|
---|
375 | for (int i = 0; i < timeSamplesArray.length; i++) {
|
---|
376 | timeSamplesArray[i] = ((double) i) / ((double) n);
|
---|
377 | }
|
---|
378 | }
|
---|
379 | timeSamples = new Matrix(timeSamplesArray, 1);
|
---|
380 | }
|
---|
381 |
|
---|
382 | private double limitConcession(double targetUtility) {
|
---|
383 | double limit = 1.0 - ((maxOfferedUtility - minOfferedUtility) + 0.1);
|
---|
384 | if (limit > targetUtility) {
|
---|
385 | return limit;
|
---|
386 | }
|
---|
387 | return targetUtility;
|
---|
388 | }
|
---|
389 |
|
---|
390 | /**
|
---|
391 | * Generate an n-by-m matrix representing the effect of the discounting
|
---|
392 | * factor for a given utility-time combination. The combinations are given
|
---|
393 | * by the time and utility samples stored in timeSamples and utilitySamples
|
---|
394 | * respectively.
|
---|
395 | *
|
---|
396 | * @param discountingFactor
|
---|
397 | * The discounting factor, in the range (0, 1].
|
---|
398 | * @return An n-by-m matrix representing the discounted utilities.
|
---|
399 | */
|
---|
400 | private Matrix generateDiscountingFunction(double discountingFactor) {
|
---|
401 | double[] discountingSamples = timeSamples.getRowPackedCopy();
|
---|
402 | double[][] m = new double[utilitySamples.getRowDimension()][timeSamples.getColumnDimension()];
|
---|
403 | for (int i = 0; i < m.length; i++) {
|
---|
404 | for (int j = 0; j < m[i].length; j++) {
|
---|
405 | m[i][j] = Math.pow(discountingFactor, discountingSamples[j]);
|
---|
406 | }
|
---|
407 | }
|
---|
408 | return new Matrix(m);
|
---|
409 | }
|
---|
410 |
|
---|
411 | /**
|
---|
412 | * Generate an (n-1)-by-m matrix representing the probability of acceptance
|
---|
413 | * for a given utility-time combination. The combinations are given by the
|
---|
414 | * time and utility samples stored in timeSamples and utilitySamples
|
---|
415 | * respectively.
|
---|
416 | *
|
---|
417 | * @param mean
|
---|
418 | * The means, at each of the sample time points.
|
---|
419 | * @param variance
|
---|
420 | * The variances, at each of the sample time points.
|
---|
421 | * @param time
|
---|
422 | * The current time, in the range [0, 1].
|
---|
423 | * @return An (n-1)-by-m matrix representing the probability of acceptance.
|
---|
424 | */
|
---|
425 | private Pair<Matrix, Matrix> generateProbabilityAccept(Matrix mean, Matrix variance, double time) {
|
---|
426 | int i = 0;
|
---|
427 | for (; i < timeSamples.getColumnDimension(); i++) {
|
---|
428 | if (timeSamples.get(0, i) > time)
|
---|
429 | break;
|
---|
430 | }
|
---|
431 | Matrix cumulativeAccept = new Matrix(utilitySamples.getRowDimension(), timeSamples.getColumnDimension(), 0);
|
---|
432 | Matrix probabilityAccept = new Matrix(utilitySamples.getRowDimension(), timeSamples.getColumnDimension(), 0);
|
---|
433 |
|
---|
434 | double interval = 1.0 / utilitySamples.getRowDimension();
|
---|
435 |
|
---|
436 | for (; i < timeSamples.getColumnDimension(); i++) {
|
---|
437 | double s = Math.sqrt(2 * variance.get(i, 0));
|
---|
438 | double m = mean.get(i, 0);
|
---|
439 |
|
---|
440 | double minp = (1.0 - (0.5 * (1 + erf((utilitySamples.get(0, 0) + (interval / 2.0) - m) / s))));
|
---|
441 | double maxp = (1.0 - (0.5 * (1
|
---|
442 | + erf((utilitySamples.get(utilitySamples.getRowDimension() - 1, 0) - (interval / 2.0) - m) / s))));
|
---|
443 |
|
---|
444 | for (int j = 0; j < utilitySamples.getRowDimension(); j++) {
|
---|
445 | double utility = utilitySamples.get(j, 0);
|
---|
446 | double p = (1.0 - (0.5 * (1 + erf((utility - m) / s))));
|
---|
447 | double p1 = (1.0 - (0.5 * (1 + erf((utility - (interval / 2.0) - m) / s))));
|
---|
448 | double p2 = (1.0 - (0.5 * (1 + erf((utility + (interval / 2.0) - m) / s))));
|
---|
449 |
|
---|
450 | cumulativeAccept.set(j, i, (p - minp) / (maxp - minp));
|
---|
451 | probabilityAccept.set(j, i, (p1 - p2) / (maxp - minp));
|
---|
452 | }
|
---|
453 | }
|
---|
454 | return new Pair<Matrix, Matrix>(probabilityAccept, cumulativeAccept);
|
---|
455 | }
|
---|
456 |
|
---|
457 | /**
|
---|
458 | * Wrapper for the erf function.
|
---|
459 | *
|
---|
460 | * @param x
|
---|
461 | * @return
|
---|
462 | */
|
---|
463 | private double erf(double x) {
|
---|
464 | if (x > 6)
|
---|
465 | return 1;
|
---|
466 | if (x < -6)
|
---|
467 | return -1;
|
---|
468 | try {
|
---|
469 | double d = Erf.erf(x);
|
---|
470 | if (d > 1)
|
---|
471 | return 1;
|
---|
472 | if (d < -1)
|
---|
473 | return -1;
|
---|
474 | return d;
|
---|
475 | } catch (MaxIterationsExceededException e) {
|
---|
476 | if (x > 0)
|
---|
477 | return 1;
|
---|
478 | else
|
---|
479 | return -1;
|
---|
480 | } catch (MathException e) {
|
---|
481 | e.printStackTrace();
|
---|
482 | return 0;
|
---|
483 | }
|
---|
484 | }
|
---|
485 |
|
---|
486 | /**
|
---|
487 | * Generate an n-by-m matrix representing the risk based utility for a given
|
---|
488 | * utility-time combination. The combinations are given by the time and
|
---|
489 | * utility samples stored in timeSamples and utilitySamples
|
---|
490 | *
|
---|
491 | * @param riskParameter
|
---|
492 | * The risk parameter.
|
---|
493 | * @return an n-by-m matrix representing the risk based utility.
|
---|
494 | */
|
---|
495 | protected Matrix generateRiskFunction(double riskParameter) {
|
---|
496 | double mmin = generateRiskFunction(riskParameter, 0.0);
|
---|
497 | double mmax = generateRiskFunction(riskParameter, 1.0);
|
---|
498 | double range = mmax - mmin;
|
---|
499 |
|
---|
500 | double[] riskSamples = utilitySamples.getColumnPackedCopy();
|
---|
501 | double[][] m = new double[utilitySamples.getRowDimension()][timeSamples.getColumnDimension()];
|
---|
502 | for (int i = 0; i < m.length; i++) {
|
---|
503 | double val;
|
---|
504 | if (range == 0) {
|
---|
505 | val = riskSamples[i];
|
---|
506 | } else {
|
---|
507 | val = (generateRiskFunction(riskParameter, riskSamples[i]) - mmin) / range;
|
---|
508 | }
|
---|
509 | for (int j = 0; j < m[i].length; j++) {
|
---|
510 | m[i][j] = val;
|
---|
511 | }
|
---|
512 | }
|
---|
513 | return new Matrix(m);
|
---|
514 | }
|
---|
515 |
|
---|
516 | /**
|
---|
517 | * Generate the risk based utility for a given actual utility.
|
---|
518 | *
|
---|
519 | * @param riskParameter
|
---|
520 | * The risk parameter.
|
---|
521 | * @param utility
|
---|
522 | * The actual utility to calculate the risk based utility from.
|
---|
523 | * @return the risk based utility.
|
---|
524 | */
|
---|
525 | protected double generateRiskFunction(double riskParameter, double utility) {
|
---|
526 | return Math.pow(utility, riskParameter);
|
---|
527 | }
|
---|
528 |
|
---|
529 | /**
|
---|
530 | * Get a pair representing the time and utility value of the expected best
|
---|
531 | * agreement.
|
---|
532 | *
|
---|
533 | * @param expectedValues
|
---|
534 | * A matrix of expected utility values at the sampled time and
|
---|
535 | * utilities given by timeSamples and utilitySamples
|
---|
536 | * respectively.
|
---|
537 | * @param time
|
---|
538 | * The current time.
|
---|
539 | * @return a pair representing the time and utility value of the expected
|
---|
540 | * best agreement.
|
---|
541 | */
|
---|
542 | private Pair<Double, Double> getExpectedBestAgreement(Matrix probabilityExpectedValues,
|
---|
543 | Matrix cumulativeExpectedValues, double time) {
|
---|
544 | Matrix probabilityFutureExpectedValues = getFutureExpectedValues(probabilityExpectedValues, time);
|
---|
545 | Matrix cumulativeFutureExpectedValues = getFutureExpectedValues(cumulativeExpectedValues, time);
|
---|
546 |
|
---|
547 | double[][] probabilityFutureExpectedValuesArray = probabilityFutureExpectedValues.getArray();
|
---|
548 | double[][] cumulativeFutureExpectedValuesArray = cumulativeFutureExpectedValues.getArray();
|
---|
549 |
|
---|
550 | Double bestX = null;
|
---|
551 | Double bestY = null;
|
---|
552 |
|
---|
553 | double[] colSums = new double[probabilityFutureExpectedValuesArray[0].length];
|
---|
554 | double bestColSum = 0;
|
---|
555 | int bestCol = 0;
|
---|
556 |
|
---|
557 | for (int x = 0; x < probabilityFutureExpectedValuesArray[0].length; x++) {
|
---|
558 | colSums[x] = 0;
|
---|
559 | for (int y = 0; y < probabilityFutureExpectedValuesArray.length; y++) {
|
---|
560 | colSums[x] += probabilityFutureExpectedValuesArray[y][x];
|
---|
561 | }
|
---|
562 |
|
---|
563 | if (colSums[x] >= bestColSum) {
|
---|
564 | bestColSum = colSums[x];
|
---|
565 | bestCol = x;
|
---|
566 | }
|
---|
567 | }
|
---|
568 |
|
---|
569 | int bestRow = 0;
|
---|
570 | double bestRowValue = 0;
|
---|
571 |
|
---|
572 | for (int y = 0; y < cumulativeFutureExpectedValuesArray.length; y++) {
|
---|
573 | double expectedValue = cumulativeFutureExpectedValuesArray[y][bestCol];
|
---|
574 | if (expectedValue > bestRowValue) {
|
---|
575 | bestRowValue = expectedValue;
|
---|
576 | bestRow = y;
|
---|
577 | }
|
---|
578 | }
|
---|
579 |
|
---|
580 | bestX = timeSamples.get(0, bestCol + probabilityExpectedValues.getColumnDimension()
|
---|
581 | - probabilityFutureExpectedValues.getColumnDimension());
|
---|
582 | bestY = utilitySamples.get(bestRow, 0);
|
---|
583 |
|
---|
584 | return new Pair<Double, Double>(bestX, bestY);
|
---|
585 | }
|
---|
586 |
|
---|
587 | /**
|
---|
588 | * Get a matrix of expected utility values at the sampled time and utilities
|
---|
589 | * given by timeSamples and utilitySamples, for times in the future.
|
---|
590 | *
|
---|
591 | * @param expectedValues
|
---|
592 | * A matrix of expected utility values at the sampled time and
|
---|
593 | * utilities given by timeSamples and utilitySamples
|
---|
594 | * respectively.
|
---|
595 | * @param time
|
---|
596 | * The current time.
|
---|
597 | * @return a matrix of expected utility values for future time.
|
---|
598 | */
|
---|
599 | private Matrix getFutureExpectedValues(Matrix expectedValues, double time) {
|
---|
600 | int i = 0;
|
---|
601 | for (; i < timeSamples.getColumnDimension(); i++) {
|
---|
602 | if (timeSamples.get(0, i) > time)
|
---|
603 | break;
|
---|
604 | }
|
---|
605 | return expectedValues.getMatrix(0, expectedValues.getRowDimension() - 1, i,
|
---|
606 | expectedValues.getColumnDimension() - 1);
|
---|
607 | }
|
---|
608 |
|
---|
609 | /**
|
---|
610 | * Get all of the bids in a utility range.
|
---|
611 | *
|
---|
612 | * @param lowerBound
|
---|
613 | * The minimum utility level of the bids.
|
---|
614 | * @param upperBound
|
---|
615 | * The maximum utility level of the bids.
|
---|
616 | * @return all of the bids in a utility range.
|
---|
617 | * @throws Exception
|
---|
618 | */
|
---|
619 | private ArrayList<Bid> getBidsInRange(double lowerBound, double upperBound) throws Exception {
|
---|
620 | ArrayList<Bid> bidsInRange = new ArrayList<Bid>();
|
---|
621 | BidIterator iter = new BidIterator(utilitySpace.getDomain());
|
---|
622 | while (iter.hasNext()) {
|
---|
623 | Bid tmpBid = iter.next();
|
---|
624 | double util = 0;
|
---|
625 | try {
|
---|
626 | util = utilitySpace.getUtility(tmpBid);
|
---|
627 | if (util >= lowerBound && util <= upperBound)
|
---|
628 | bidsInRange.add(tmpBid);
|
---|
629 | } catch (Exception e) {
|
---|
630 | e.printStackTrace();
|
---|
631 | }
|
---|
632 | }
|
---|
633 |
|
---|
634 | return bidsInRange;
|
---|
635 | }
|
---|
636 |
|
---|
637 | /**
|
---|
638 | * Get a random bid in a given utility range.
|
---|
639 | *
|
---|
640 | * @param lowerBound
|
---|
641 | * The lower bound on utility.
|
---|
642 | * @param upperBound
|
---|
643 | * The upper bound on utility.
|
---|
644 | * @return a random bid in a given utility range.
|
---|
645 | * @throws Exception
|
---|
646 | */
|
---|
647 | protected Bid getRandomBidInRange(double lowerBound, double upperBound) throws Exception {
|
---|
648 | ArrayList<Bid> bidsInRange = getBidsInRange(lowerBound, upperBound);
|
---|
649 |
|
---|
650 | int index = (new Random()).nextInt(bidsInRange.size() - 1);
|
---|
651 |
|
---|
652 | return bidsInRange.get(index);
|
---|
653 | }
|
---|
654 |
|
---|
655 | public double adjustDiscountFactor(double discountFactor) {
|
---|
656 | return discountFactor;
|
---|
657 | }
|
---|
658 |
|
---|
659 | @Override
|
---|
660 | public String getName() {
|
---|
661 | return "2012 - IAMHaggler";
|
---|
662 | }
|
---|
663 | }
|
---|