[75] | 1 | import logging
|
---|
| 2 | from random import randint
|
---|
| 3 | from time import time
|
---|
| 4 | from typing import cast
|
---|
| 5 |
|
---|
| 6 | import pandas as pd
|
---|
| 7 | from geniusweb.actions.Accept import Accept
|
---|
| 8 | from geniusweb.actions.Action import Action
|
---|
| 9 | from geniusweb.actions.Offer import Offer
|
---|
| 10 | from geniusweb.actions.PartyId import PartyId
|
---|
| 11 | from geniusweb.bidspace.AllBidsList import AllBidsList
|
---|
| 12 | from geniusweb.inform.ActionDone import ActionDone
|
---|
| 13 | from geniusweb.inform.Finished import Finished
|
---|
| 14 | from geniusweb.inform.Inform import Inform
|
---|
| 15 | from geniusweb.inform.Settings import Settings
|
---|
| 16 | from geniusweb.inform.YourTurn import YourTurn
|
---|
| 17 | from geniusweb.issuevalue.Bid import Bid
|
---|
| 18 | from geniusweb.issuevalue.Domain import Domain
|
---|
| 19 | from geniusweb.party.Capabilities import Capabilities
|
---|
| 20 | from geniusweb.party.DefaultParty import DefaultParty
|
---|
| 21 | from geniusweb.profile.utilityspace.LinearAdditiveUtilitySpace import (
|
---|
| 22 | LinearAdditiveUtilitySpace,
|
---|
| 23 | )
|
---|
| 24 | from geniusweb.profileconnection.ProfileConnectionFactory import (
|
---|
| 25 | ProfileConnectionFactory,
|
---|
| 26 | )
|
---|
| 27 | from geniusweb.progress.ProgressTime import ProgressTime
|
---|
| 28 | from geniusweb.references.Parameters import Parameters
|
---|
| 29 | from tudelft_utilities_logging.ReportToLogger import ReportToLogger
|
---|
| 30 |
|
---|
| 31 | from agents.template_agent.utils.opponent_model import OpponentModel
|
---|
| 32 |
|
---|
| 33 | # our imports
|
---|
| 34 | import numpy as np
|
---|
| 35 | from sklearn import tree
|
---|
| 36 | from sklearn.preprocessing import label_binarize
|
---|
| 37 | import random
|
---|
| 38 |
|
---|
| 39 |
|
---|
| 40 | class GEAAgent(DefaultParty):
|
---|
| 41 | """
|
---|
| 42 | Template of a Python geniusweb agent.
|
---|
| 43 | """
|
---|
| 44 |
|
---|
| 45 | def __init__(self):
|
---|
| 46 | super().__init__()
|
---|
| 47 | self.logger: ReportToLogger = self.getReporter()
|
---|
| 48 |
|
---|
| 49 | self.domain: Domain = None
|
---|
| 50 | self.parameters: Parameters = None
|
---|
| 51 | self.profile: LinearAdditiveUtilitySpace = None
|
---|
| 52 | self.progress: ProgressTime = None
|
---|
| 53 | self.me: PartyId = None
|
---|
| 54 | self.other: str = None
|
---|
| 55 | self.settings: Settings = None
|
---|
| 56 | self.storage_dir: str = None
|
---|
| 57 |
|
---|
| 58 | self.last_received_bid: Bid = None
|
---|
| 59 | self.opponent_model: OpponentModel = None
|
---|
| 60 | self.logger.log(logging.INFO, "party is initialized")
|
---|
| 61 |
|
---|
| 62 | # our parameters
|
---|
| 63 | # collect negitioation data
|
---|
| 64 | self.dataX = []
|
---|
| 65 | self.dataY = []
|
---|
| 66 | self.data_len = 0
|
---|
| 67 | self.issue_encoder = {}
|
---|
| 68 |
|
---|
| 69 | # decision tree and weights
|
---|
| 70 | self.decision_model = None
|
---|
| 71 | self.tree_depth = 20
|
---|
| 72 | self.orig_opponent_agree_weight = 0.15
|
---|
| 73 | self.opponent_agree_weight = self.orig_opponent_agree_weight
|
---|
| 74 | self.accept_threshold = 0.85 # for heuristic function, not utility.
|
---|
| 75 |
|
---|
| 76 | # define
|
---|
| 77 | self.OPPONENT_ACCEPT = 1
|
---|
| 78 | self.OPPONENT_REJECT = -1
|
---|
| 79 |
|
---|
| 80 | # bid dictionaries
|
---|
| 81 | self.bid_values = {}
|
---|
| 82 | self.all_issue_values = {}
|
---|
| 83 |
|
---|
| 84 | # bid lookup indices
|
---|
| 85 | # self.lower_threshold = 0
|
---|
| 86 | # self.higher_threshold = 200
|
---|
| 87 |
|
---|
| 88 | self.last_bid_utility = 0
|
---|
| 89 | self.debug_offer = 0
|
---|
| 90 |
|
---|
| 91 | self.randomn = random.uniform(0, 1)
|
---|
| 92 |
|
---|
| 93 | def notifyChange(self, data: Inform):
|
---|
| 94 | """MUST BE IMPLEMENTED
|
---|
| 95 | This is the entry point of all interaction with your agent after is has been initialised.
|
---|
| 96 | How to handle the received data is based on its class type.
|
---|
| 97 |
|
---|
| 98 | Args:
|
---|
| 99 | info (Inform): Contains either a request for action or information.
|
---|
| 100 | """
|
---|
| 101 |
|
---|
| 102 | # a Settings message is the first message that will be send to your
|
---|
| 103 | # agent containing all the information about the negotiation session.
|
---|
| 104 | if isinstance(data, Settings):
|
---|
| 105 | self.settings = cast(Settings, data)
|
---|
| 106 | self.me = self.settings.getID()
|
---|
| 107 |
|
---|
| 108 | # progress towards the deadline has to be tracked manually through the use of the Progress object
|
---|
| 109 | self.progress = self.settings.getProgress()
|
---|
| 110 |
|
---|
| 111 | self.parameters = self.settings.getParameters()
|
---|
| 112 | self.storage_dir = self.parameters.get("storage_dir")
|
---|
| 113 |
|
---|
| 114 | # the profile contains the preferences of the agent over the domain
|
---|
| 115 | profile_connection = ProfileConnectionFactory.create(
|
---|
| 116 | data.getProfile().getURI(), self.getReporter()
|
---|
| 117 | )
|
---|
| 118 | self.profile = profile_connection.getProfile()
|
---|
| 119 | self.domain = self.profile.getDomain()
|
---|
| 120 | profile_connection.close()
|
---|
| 121 |
|
---|
| 122 | # our code: init issue dictionary
|
---|
| 123 | self.init_bid_values()
|
---|
| 124 |
|
---|
| 125 | # ActionDone informs you of an action (an offer or an accept)
|
---|
| 126 | # that is performed by one of the agents (including yourself).
|
---|
| 127 | elif isinstance(data, ActionDone):
|
---|
| 128 | action = cast(ActionDone, data).getAction()
|
---|
| 129 | actor = action.getActor()
|
---|
| 130 |
|
---|
| 131 | # ignore action if it is our action
|
---|
| 132 | if actor != self.me:
|
---|
| 133 | # obtain the name of the opponent, cutting of the position ID.
|
---|
| 134 | self.other = str(actor).rsplit("_", 1)[0]
|
---|
| 135 |
|
---|
| 136 | # process action done by opponent
|
---|
| 137 | self.opponent_action(action)
|
---|
| 138 | # YourTurn notifies you that it is your turn to act
|
---|
| 139 | elif isinstance(data, YourTurn):
|
---|
| 140 | # execute a turn
|
---|
| 141 | self.my_turn()
|
---|
| 142 |
|
---|
| 143 | # Finished will be send if the negotiation has ended (through agreement or deadline)
|
---|
| 144 | elif isinstance(data, Finished):
|
---|
| 145 | self.save_data()
|
---|
| 146 | # terminate the agent MUST BE CALLED
|
---|
| 147 | self.logger.log(logging.INFO, "party is terminating:")
|
---|
| 148 | super().terminate()
|
---|
| 149 | else:
|
---|
| 150 | self.logger.log(logging.WARNING, "Ignoring unknown info " + str(data))
|
---|
| 151 |
|
---|
| 152 | def getCapabilities(self) -> Capabilities:
|
---|
| 153 | """MUST BE IMPLEMENTED
|
---|
| 154 | Method to indicate to the protocol what the capabilities of this agent are.
|
---|
| 155 | Leave it as is for the ANL 2022 competition
|
---|
| 156 |
|
---|
| 157 | Returns:
|
---|
| 158 | Capabilities: Capabilities representation class
|
---|
| 159 | """
|
---|
| 160 | return Capabilities(
|
---|
| 161 | set(["SAOP"]),
|
---|
| 162 | set(["geniusweb.profile.utilityspace.LinearAdditive"]),
|
---|
| 163 | )
|
---|
| 164 |
|
---|
| 165 | def send_action(self, action: Action):
|
---|
| 166 | """Sends an action to the opponent(s)
|
---|
| 167 |
|
---|
| 168 | Args:
|
---|
| 169 | action (Action): action of this agent
|
---|
| 170 | """
|
---|
| 171 | self.getConnection().send(action)
|
---|
| 172 |
|
---|
| 173 | # give a description of your agent
|
---|
| 174 | def getDescription(self) -> str:
|
---|
| 175 | """MUST BE IMPLEMENTED
|
---|
| 176 | Returns a description of your agent. 1 or 2 sentences.
|
---|
| 177 |
|
---|
| 178 | Returns:
|
---|
| 179 | str: Agent description
|
---|
| 180 | """
|
---|
| 181 | return "Template agent for the ANL 2022 competition"
|
---|
| 182 |
|
---|
| 183 | def opponent_action(self, action):
|
---|
| 184 | """Process an action that was received from the opponent.
|
---|
| 185 |
|
---|
| 186 | Args:
|
---|
| 187 | action (Action): action of opponent
|
---|
| 188 | """
|
---|
| 189 | # if it is an offer, set the last received bid
|
---|
| 190 | if isinstance(action, Offer):
|
---|
| 191 | # create opponent model if it was not yet initialised
|
---|
| 192 | if self.opponent_model is None:
|
---|
| 193 | self.opponent_model = OpponentModel(self.domain)
|
---|
| 194 |
|
---|
| 195 | bid = cast(Offer, action).getBid()
|
---|
| 196 |
|
---|
| 197 | # update opponent model with bid
|
---|
| 198 | self.opponent_model.update(bid)
|
---|
| 199 | # set bid as last received
|
---|
| 200 | self.last_received_bid = bid
|
---|
| 201 |
|
---|
| 202 | def my_turn(self):
|
---|
| 203 | """This method is called when it is our turn. It should decide upon an action
|
---|
| 204 | to perform and send this action to the opponent.
|
---|
| 205 | """
|
---|
| 206 |
|
---|
| 207 | # check if the last received offer is good enough
|
---|
| 208 | if self.accept_condition(self.last_received_bid):
|
---|
| 209 | # if so, accept the offer
|
---|
| 210 | action = Accept(self.me, self.last_received_bid)
|
---|
| 211 | else:
|
---|
| 212 | # if not, find a bid to propose as counter offer
|
---|
| 213 | bid = self.find_bid()
|
---|
| 214 | action = Offer(self.me, bid)
|
---|
| 215 | self.append_data_and_train_tree(bid, self.OPPONENT_REJECT)
|
---|
| 216 |
|
---|
| 217 | # send the action
|
---|
| 218 | self.send_action(action)
|
---|
| 219 |
|
---|
| 220 | def save_data(self):
|
---|
| 221 | """This method is called after the negotiation is finished. It can be used to store data
|
---|
| 222 | for learning capabilities. Note that no extensive calculations can be done within this method.
|
---|
| 223 | Taking too much time might result in your agent being killed, so use it for storage only.
|
---|
| 224 | """
|
---|
| 225 | data = "Data for learning (see README.md)"
|
---|
| 226 | with open(f"{self.storage_dir}/data.md", "w") as f:
|
---|
| 227 | f.write(data)
|
---|
| 228 |
|
---|
| 229 | ###########################################################################################
|
---|
| 230 | ################################## Example methods below ##################################
|
---|
| 231 | ###########################################################################################
|
---|
| 232 |
|
---|
| 233 | def accept_condition(self, bid: Bid) -> bool:
|
---|
| 234 | if bid is None:
|
---|
| 235 | return False
|
---|
| 236 |
|
---|
| 237 | # our code
|
---|
| 238 | # process new bid offer
|
---|
| 239 | heuristic_score = self.score_bid(bid)
|
---|
| 240 | objective_utility = self.profile.getUtility(bid)
|
---|
| 241 |
|
---|
| 242 | self.append_data_and_train_tree(bid, self.OPPONENT_ACCEPT)
|
---|
| 243 |
|
---|
| 244 | if objective_utility < self.last_bid_utility:
|
---|
| 245 | self.opponent_agree_weight *= 0.99
|
---|
| 246 | if objective_utility > self.last_bid_utility:
|
---|
| 247 | self.opponent_agree_weight /= 0.99
|
---|
| 248 |
|
---|
| 249 | if self.opponent_agree_weight == 0:
|
---|
| 250 | self.opponent_agree_weight = 0.01
|
---|
| 251 |
|
---|
| 252 | self.last_bid_utility = objective_utility
|
---|
| 253 |
|
---|
| 254 | # progress of the negotiation session between 0 and 1 (1 is deadline)
|
---|
| 255 | progress = self.progress.get(time() * 1000)
|
---|
| 256 |
|
---|
| 257 | conditions = [
|
---|
| 258 | progress > 0.95 and objective_utility > 0.4,
|
---|
| 259 | progress > 0.9 and objective_utility > 0.6,
|
---|
| 260 | objective_utility > 0.7,
|
---|
| 261 | heuristic_score >= self.accept_threshold,
|
---|
| 262 | ]
|
---|
| 263 | return any(conditions)
|
---|
| 264 |
|
---|
| 265 | def find_bid(self) -> Bid:
|
---|
| 266 | # compose a list of all possible bids
|
---|
| 267 | domain = self.profile.getDomain()
|
---|
| 268 | all_bids = AllBidsList(domain)
|
---|
| 269 |
|
---|
| 270 | best_bid_score = 0.0
|
---|
| 271 | best_bid = None
|
---|
| 272 |
|
---|
| 273 | # take 500 attempts to find a bid according to a heuristic score
|
---|
| 274 | for _ in range(500):
|
---|
| 275 | bid = all_bids.get(randint(0, all_bids.size() - 1))
|
---|
| 276 | bid_score = self.score_bid(bid)
|
---|
| 277 | if bid_score > best_bid_score:
|
---|
| 278 | best_bid_score, best_bid = bid_score, bid
|
---|
| 279 |
|
---|
| 280 | return best_bid
|
---|
| 281 |
|
---|
| 282 | def score_bid(self, bid: Bid, alpha: float = 0.95, eps: float = 0.1) -> float:
|
---|
| 283 | ''' Calculate heuristic score for a bid '''
|
---|
| 284 | progress = self.progress.get(time() * 1000)
|
---|
| 285 |
|
---|
| 286 | our_utility = float(self.profile.getUtility(bid))
|
---|
| 287 |
|
---|
| 288 | time_pressure = 1.0 - progress ** (1 / eps)
|
---|
| 289 | score = alpha * time_pressure * our_utility
|
---|
| 290 |
|
---|
| 291 | opponent_score = self.tree_predict(bid) * self.opponent_agree_weight
|
---|
| 292 | score += opponent_score
|
---|
| 293 |
|
---|
| 294 | return score
|
---|
| 295 |
|
---|
| 296 |
|
---|
| 297 | def tree_predict(self, bid: Bid) -> float:
|
---|
| 298 | ''' returns acceptance estimation for the other agent '''
|
---|
| 299 | bid_data = []
|
---|
| 300 | bid_issue_values = bid.getIssueValues()
|
---|
| 301 | domain_issues = list(bid_issue_values.keys())
|
---|
| 302 | domain_issues.sort()
|
---|
| 303 | for issue in domain_issues:
|
---|
| 304 | # encode categorical data
|
---|
| 305 | issue_encoded = label_binarize([str(bid_issue_values[issue])], classes=self.all_issue_values[issue])
|
---|
| 306 | # concat current category to X
|
---|
| 307 | bid_data.extend(issue_encoded.flatten().tolist())
|
---|
| 308 |
|
---|
| 309 | # if the tree is trained, we can use it to predict opponent reaction
|
---|
| 310 | if self.decision_model is not None:
|
---|
| 311 | tree_prediction = float(self.decision_model.predict(np.array(bid_data).reshape(1, -1)))
|
---|
| 312 | return tree_prediction
|
---|
| 313 |
|
---|
| 314 | return 0 # no knowledge
|
---|
| 315 |
|
---|
| 316 | def append_data_and_train_tree(self, bid: Bid, opponent_accept: int) -> None:
|
---|
| 317 | ''' appends new bid to negotiation history and retrain model '''
|
---|
| 318 | bid_data = []
|
---|
| 319 | bid_issue_values = bid.getIssueValues()
|
---|
| 320 | domain_issues = list(bid_issue_values.keys())
|
---|
| 321 | domain_issues.sort()
|
---|
| 322 | for issue in domain_issues:
|
---|
| 323 | # encode categorical data
|
---|
| 324 | issue_encoded = label_binarize([str(bid_issue_values[issue])], classes=self.all_issue_values[issue])
|
---|
| 325 | # concat current category to X
|
---|
| 326 | bid_data.extend(issue_encoded.flatten().tolist())
|
---|
| 327 |
|
---|
| 328 | self.data_len += 1
|
---|
| 329 | self.dataX.append(bid_data)
|
---|
| 330 | self.dataY.append(opponent_accept)
|
---|
| 331 |
|
---|
| 332 | # train tree if at least two samples were collected
|
---|
| 333 |
|
---|
| 334 | if self.data_len > 2:
|
---|
| 335 | self.decision_model = tree.DecisionTreeClassifier(criterion="entropy", max_depth=self.tree_depth)
|
---|
| 336 | self.decision_model.fit(self.dataX, self.dataY)
|
---|
| 337 |
|
---|
| 338 | def init_bid_values(self):
|
---|
| 339 | ''' must be called to binarize labels '''
|
---|
| 340 | domain = self.profile.getDomain()
|
---|
| 341 | domain_issues = domain.getIssues()
|
---|
| 342 | self.all_issue_values = {}
|
---|
| 343 | for issue in domain_issues:
|
---|
| 344 | self.all_issue_values[issue] = []
|
---|
| 345 | for value in domain.getValues(issue):
|
---|
| 346 | self.all_issue_values[issue].append(str(value))
|
---|