source: src/main/java/agents/anac/y2019/garavelagent/GaravelAgent.java@ 207

Last change on this file since 207 was 207, checked in by Tim Baarslag, 5 years ago

One to many negotiation classes
  • Property svn:executable set to *
File size: 13.2 KB
Line 
1package agents.anac.y2019.garavelagent;
2
3import java.util.List;
4
5import java.util.ArrayList;
6import java.util.Arrays;
7import java.util.Comparator;
8import java.util.Random;
9
10import agents.org.apache.commons.math.stat.regression.OLSMultipleLinearRegression;
11import genius.core.AgentID;
12import genius.core.Bid;
13import genius.core.actions.Accept;
14import genius.core.actions.Action;
15import genius.core.actions.Offer;
16import genius.core.issue.Issue;
17import genius.core.issue.IssueDiscrete;
18import genius.core.issue.ValueDiscrete;
19import genius.core.parties.AbstractNegotiationParty;
20import genius.core.parties.NegotiationInfo;
21import genius.core.uncertainty.AdditiveUtilitySpaceFactory;
22import genius.core.uncertainty.BidRanking;
23import genius.core.utility.AbstractUtilitySpace;
24import javafx.util.Pair;
25
26/**
27 * This is your negotiation party.
28 */
29
30public class GaravelAgent extends AbstractNegotiationParty {
31
32 private Bid currentBid = null;
33 private Bid beforeBid = null;
34 private AbstractUtilitySpace utilitySpace = null;
35 private double[][] opponentModelValue;
36 private double[] opponentModelIssue;
37 private int numberOfIssues;
38 private ArrayList<Bid> opponentBids;
39 private int bidCount = 0;
40 double[] estimateUtilities;
41 double[] sortedUtils;
42
43 List<List<ValueDiscrete>> allIssues = new ArrayList<>();
44 int countAll;
45 double[][] omValueNormalized;
46 OLSMultipleLinearRegression regression;
47 ArrayList<Bid> allBidsList;
48 double[] utilities;
49 List<List<String>> allIssuesAsString;
50 List<List<String>> allPossibleBids;
51 List<Pair<List<String>, Double>> ourUtilityModel;
52 List<Pair<List<String>, Double>> opponentUtilityModel;
53 NegotiationInfo info_;
54
55 ArrayList<Bid> sortedList;
56 double[] allBidPredictions;
57 Bid maxBid = null;
58 List<List<String>> optimalBids;
59 Boolean isRegressionPossible = true;
60 List<Bid> bidOrder;
61
62 @Override
63 public void init(NegotiationInfo info) {
64
65 super.init(info);
66
67 try {
68 List<Issue> issues = info.getUserModel().getDomain().getIssues();
69 utils.getIssueDiscrete(issues, allIssues);
70 allIssuesAsString = issuesAsString();
71 allPossibleBids = utils.generateAllPossibleBids(allIssuesAsString, 0);
72 utils.reverse(allPossibleBids);
73 countAll = utils.getIssueCount(allIssues);
74 info_ = info;
75
76 bidOrder = info.getUserModel().getBidRanking().getBidOrder();
77 AdditiveUtilitySpaceFactory factory = new AdditiveUtilitySpaceFactory(getDomain());
78 BidRanking bidRanking = userModel.getBidRanking();
79 factory.estimateUsingBidRanks(bidRanking);
80 utilitySpace = getUtilitySpace();
81
82 double[][] oneHotEncoded = utils.encodeBids(bidOrder, countAll, allIssues);
83 double[][] oneHotEncodedAll = utils.encodeListOfStrings(allPossibleBids, countAll, allIssues);
84
85 utilities = new double[bidOrder.size()];
86 double highBid = info.getUserModel().getBidRanking().getHighUtility();
87 double lowBid = info.getUserModel().getBidRanking().getLowUtility();
88 utilities[0] = lowBid;
89 utilities[utilities.length - 1] = highBid;
90
91 double delta = highBid - lowBid;
92 double decrementAmount = delta / (utilities.length - 1);
93
94
95 for (int i = 1; i < utilities.length - 1; i++) {
96 utilities[i] = utilities[i - 1] + decrementAmount;
97 }
98
99 regression = new OLSMultipleLinearRegression();
100 regression.newSampleData(utilities, oneHotEncoded);
101
102 allBidPredictions = new double[oneHotEncodedAll.length];
103
104 for (int i = 0; i < oneHotEncodedAll.length; i++) {
105 allBidPredictions[i] = utils.predict(regression, oneHotEncodedAll[i]);
106 }
107
108 double max = Arrays.stream(allBidPredictions).max().getAsDouble();
109
110 for (int i = 0; i < oneHotEncodedAll.length; i++) {
111 if (allBidPredictions[i] > 0.9) {
112 allBidPredictions[i] = utils.scale(allBidPredictions[i], 0.9, max);
113 }
114 }
115
116 Bid sampleBid = generateRandomBid();
117 numberOfIssues = sampleBid.getIssues().size();
118 // Array to keep issueWeights we plan to sample
119 opponentModelIssue = new double[numberOfIssues];
120
121 // Array to keep valueWeights we plan to sample
122 opponentModelValue = new double[numberOfIssues][];
123 opponentBids = new ArrayList<>();
124
125 // Creating the 2d array by initializing non-fixed size rows
126 for (int i = 0; i < numberOfIssues; i++) {
127 opponentModelIssue[i] = (double) 1 / numberOfIssues;
128 IssueDiscrete issueDiscrete = (IssueDiscrete) sampleBid.getIssues().get(i);
129 opponentModelValue[i] = new double[issueDiscrete.getValues().size()];
130 }
131
132 allBidsList = new ArrayList<>();
133 for (int i = 0; i < allPossibleBids.size(); i++) {
134 allBidsList.add(utils.asBid(info.getUserModel().getDomain(), convertToStringArray(allPossibleBids.get(i).toArray())));
135 }
136
137 final List<Bid> allBidsCopy = allBidsList;
138 sortedList = new ArrayList<>(allBidsCopy);
139 sortedList.sort(Comparator.comparing(s -> allBidPredictions[allBidsCopy.indexOf(s)]));
140 sortedUtils = Arrays.stream(allBidPredictions).sorted().toArray();
141
142 ourUtilityModel = new ArrayList<>();
143 for (int i = 0; i < sortedList.size(); i++) {
144 ourUtilityModel.add(new Pair<>(utils.bidToListOfString(sortedList.get(i)), sortedUtils[i]));
145 }
146
147 } catch (Exception e) {
148 isRegressionPossible = false;
149 }
150
151 }
152
153 private static String[] convertToStringArray(Object[] array) {
154 String[] result = new String[array.length];
155 for (int i = 0; i < array.length; i++) {
156 result[i] = array[i].toString();
157 }
158 return result;
159 }
160
161 private List<List<String>> issuesAsString() {
162 List<List<String>> allIssuesAsString = new ArrayList<>();
163 for (int i = 0; i < allIssues.size(); i++) {
164 List<String> current = new ArrayList<>();
165 for (int j = 0; j < allIssues.get(i).size(); j++) {
166 current.add(allIssues.get(i).get(j).toString());
167 }
168 allIssuesAsString.add(current);
169 }
170 return allIssuesAsString;
171 }
172
173 @Override
174 public Action chooseAction(List<Class<? extends Action>> validActions) {
175
176 if (isRegressionPossible) {
177 try {
178
179 if (maxBid == null)
180 maxBid = currentBid;
181
182 if (allBidsList.indexOf(currentBid) < allBidsList.indexOf(maxBid))
183 maxBid = currentBid;
184
185 if (currentBid == null || !validActions.contains(Accept.class)) {
186 return new Offer(getPartyId(), getUtilitySpace().getMaxUtilityBid());
187 }
188
189 if (timeline.getCurrentTime() == 998) {
190 return new Offer(getPartyId(), getUtilitySpace().getMaxUtilityBid());
191 } else if (timeline.getCurrentTime() == 999) {
192 return new Offer(getPartyId(), getUtilitySpace().getMaxUtilityBid());
193 } else if (timeline.getCurrentTime() == 1000) {
194 if (0.84 <= utils.predict(regression, utils.encodeBid(maxBid, countAll, allIssues))) {
195 return new Offer(getPartyId(), maxBid);
196 } else {
197 try {
198 Random r = new Random();
199 Bid lastBid = utils.asBid(info_.getUserModel().getDomain(), toStringArray(optimalBids.get(r.nextInt(optimalBids.size()))));
200 if (0.70 <= utils.predict(regression, utils.encodeBid(lastBid, countAll, allIssues))) {
201 return new Offer(getPartyId(), lastBid);
202 } else {
203 return new Offer(getPartyId(), getUtilitySpace().getMaxUtilityBid());
204 }
205 } catch (Exception e) {
206 return new Offer(getPartyId(), getUtilitySpace().getMaxUtilityBid());
207 }
208
209 }
210 }
211
212 if (0.92 <= utils.predict(regression, utils.encodeBid(currentBid, countAll, allIssues))) {
213 System.out.println("accepted");
214 return new Accept(getPartyId(), currentBid);
215 } else {
216 try {
217 if (timeline.getCurrentTime() >= 200) {
218 Random r = new Random();
219 return new Offer(getPartyId(), utils.asBid(info_.getUserModel().getDomain(), toStringArray(optimalBids.get(r.nextInt(optimalBids.size())))));
220 }
221 } catch (Exception e) {
222 System.out.println("freq failed");
223 return new Offer(getPartyId(), getUtilitySpace().getMaxUtilityBid());
224 }
225 return new Offer(getPartyId(), getUtilitySpace().getMaxUtilityBid());
226 }
227
228 } catch (Exception e) {
229 e.printStackTrace();
230 return new Offer(getPartyId(), generateRandomBid());
231 } finally {
232 if (currentBid != null) {
233 beforeBid = currentBid;
234 }
235 }
236 } else {
237 if (timeline.getCurrentTime() == 995) {
238 return new Offer(getPartyId(), bidOrder.get(bidOrder.size() - 1));
239 }else{
240 return new Offer(getPartyId(), bidOrder.get(bidOrder.size() - 1));
241 }
242 }
243 }
244
245 private void updateWeights() {
246 for (int i = 0; i < numberOfIssues; i++) {
247 ValueDiscrete currentBidValue = (ValueDiscrete) (currentBid.getValue(i + 1));
248 int currentBidValueIndex = utils.getIndexOfValueInIssue(i, currentBidValue.getValue(), currentBid);
249
250 ValueDiscrete beforeBidValue = (ValueDiscrete) (beforeBid.getValue(i + 1));
251 int beforeBidValueIndex = utils.getIndexOfValueInIssue(i, beforeBidValue.getValue(), currentBid);
252
253 if (currentBidValueIndex == beforeBidValueIndex) {
254 opponentModelIssue[i] += ((double) 1 / numberOfIssues);
255 }
256 opponentModelValue[i][currentBidValueIndex] += 1;
257 }
258 }
259
260 // If we hold 2 bids, update the weights with it!
261 @Override
262 public void receiveMessage(AgentID sender, Action action) {
263
264 try {
265 super.receiveMessage(sender, action);
266 if (action instanceof Offer) {
267
268 currentBid = ((Offer) action).getBid();
269 opponentBids.add(currentBid);
270 if (!currentBid.equals(beforeBid) && beforeBid != null) {
271 updateWeights();
272 updateOMValues();
273
274 if (bidCount % 50 == 0) {
275 //updateOpponentModel();
276 estimateUtilities = estimateOpUtil(allPossibleBids, omValueNormalized);
277 opponentUtilityModel = utils.frequencyModelling(opponentBids, allIssuesAsString, opponentModelValue);
278
279 //optimal bids to offer are stored below
280 optimalBids = utils.getOptimalBids(allPossibleBids, utils.mostWanted, regression, info_.getUserModel().getDomain(), countAll, allIssues);
281 }
282 }
283 bidCount++;
284 }
285 } catch (Exception e) {
286 //e.printStackTrace();
287 }
288 }
289
290 private String[] toStringArray(List<String> input) {
291 String[] result = new String[input.size()];
292 for (int i = 0; i < input.size(); i++) {
293 result[i] = input.get(i);
294 }
295 return result;
296 }
297
298 private void updateOMValues() {
299
300 double sum = 0;
301 for (int i = 0; i < opponentModelValue[0].length; i++) {
302 sum += opponentModelValue[0][i];
303 }
304
305 omValueNormalized = new double[opponentModelValue.length][opponentModelValue[0].length];
306 for (int i = 0; i < omValueNormalized.length; i++)
307 omValueNormalized[i] = Arrays.copyOf(opponentModelValue[i], opponentModelValue[i].length);
308
309 for (int i = 0; i < opponentModelValue.length; i++) {
310 for (int j = 0; j < opponentModelValue[i].length; j++) {
311 omValueNormalized[i][j] = opponentModelValue[i][j] / sum;
312 }
313 }
314
315 }
316
317 private double[] estimateOpUtil(List<List<String>> allPossibleBids, double[][] omValueNormalized) {
318 double[] result = new double[allPossibleBids.size()];
319 for (int i = 0; i < allPossibleBids.size(); i++) {
320 for (int j = 0; j < allIssuesAsString.size(); j++) {
321 for (int k = 0; k < allIssuesAsString.get(j).size(); k++) {
322 if (allPossibleBids.get(i).get(j).equals(allIssuesAsString.get(j).get(k))) {
323 result[i] += omValueNormalized[j][k];
324 }
325 }
326 }
327 }
328 return result;
329 }
330
331 @Override
332 public String getDescription() {
333 return "mezgit_soft";
334 }
335
336}
Note: See TracBrowser for help on using the repository browser.