1 | import logging
|
---|
2 | import time
|
---|
3 | import random
|
---|
4 | from random import randint, choices
|
---|
5 | from typing import cast
|
---|
6 |
|
---|
7 | import geniusweb.opponentmodel.FrequencyOpponentModel as freq_opp_mod
|
---|
8 | import numpy as np
|
---|
9 | from geniusweb.actions.Accept import Accept
|
---|
10 | from geniusweb.actions.Action import Action
|
---|
11 | from geniusweb.actions.Offer import Offer
|
---|
12 | from geniusweb.bidspace.AllBidsList import AllBidsList
|
---|
13 | from geniusweb.inform.ActionDone import ActionDone
|
---|
14 | from geniusweb.inform.Finished import Finished
|
---|
15 | from geniusweb.inform.Inform import Inform
|
---|
16 | from geniusweb.inform.Settings import Settings
|
---|
17 | from geniusweb.inform.YourTurn import YourTurn
|
---|
18 | from geniusweb.issuevalue.Bid import Bid
|
---|
19 | from geniusweb.party.Capabilities import Capabilities
|
---|
20 | from geniusweb.party.DefaultParty import DefaultParty
|
---|
21 | from geniusweb.profileconnection.ProfileConnectionFactory import (
|
---|
22 | ProfileConnectionFactory,
|
---|
23 | )
|
---|
24 | from geniusweb.progress.Progress import Progress
|
---|
25 | from .acceptance_strategy import AcceptanceStrategy
|
---|
26 | from geniusweb.progress.ProgressRounds import ProgressRounds
|
---|
27 | from tudelft_utilities_logging.Reporter import Reporter
|
---|
28 |
|
---|
29 |
|
---|
30 | # A custom agent that combines different strategies and changes between them based on time
|
---|
31 | # At first the agent enters an exploration phase where it acts as a very strict random walker
|
---|
32 | # After the exploration phase the agent starts behaving like the Agreeable agent, picking bids based on minimum
|
---|
33 | # utility and roulette selection based on social welfare
|
---|
34 | # After that, if the agents still did not find an agreement, the agent will start looking for the best nash product
|
---|
35 | # Lastly the agent will start sending bids that it already received, maximizing its utility
|
---|
36 | class Agent18(DefaultParty):
|
---|
37 | """
|
---|
38 | -- Shreker --
|
---|
39 | The Shreker agent is an agent that changes its strategy depending on the time in the following order:
|
---|
40 | - Random walker: initially explores opponent utility space while prevent opponent from getting our best bids
|
---|
41 | - Agreeable: agent by Sahar Mirzayi from ANAC 2018; offers the highest utility bid that concedes on one issue
|
---|
42 | from the offer
|
---|
43 | - Social welfare: late into the negotiation optimizes social welfare if opponent still has not conceded much
|
---|
44 | - Received bids: very late into the negotiation return one of the best bids out of the 20 last received bids
|
---|
45 | """
|
---|
46 |
|
---|
47 | def __init__(self, reporter: Reporter = None):
|
---|
48 | super().__init__(reporter)
|
---|
49 | self.getReporter().log(logging.INFO, "party is initialized")
|
---|
50 | self._profile = None
|
---|
51 | # Stores the last received bid
|
---|
52 | self._last_received_bid: Bid = None
|
---|
53 | # List of all received bids
|
---|
54 | self._received_bids: list[Bid] = []
|
---|
55 | # Stores the last sent bid
|
---|
56 | self._last_sent_bid = None
|
---|
57 | # Stores the best utility stored so far
|
---|
58 | self._best_received_utility = 0.0
|
---|
59 | # Stores all the thresholds used throughout the agent
|
---|
60 | # 0 -> Threshold for acceptance strategy
|
---|
61 | # 1 -> Threshold for random walker | RandomWalker
|
---|
62 | # 2 -> Minimum target utility | Agreeable
|
---|
63 | # 3 -> Factor of the time dependent utility | Agreeable
|
---|
64 | # 4,5,6 -> Time splits for changing strategies
|
---|
65 | self.thresholds: list[float] = [0.99, 0.980278280105376, 0.9586147509907781, 3.846489410609955,
|
---|
66 | 0.5702511194471804, 0.8702511194471804, 0.99]
|
---|
67 | # Ranges for the thresholds for optimization purposes
|
---|
68 | self.threshold_checks = [[0.8, 1], [0.7, 1], [0.7, 1], [2, 4],
|
---|
69 | [0.3, 0.7], [0.7, 0.9], [0.9, 1]]
|
---|
70 |
|
---|
71 | def notifyChange(self, info: Inform):
|
---|
72 | """This is the entry point of all interaction with your agent after is has been initialised.
|
---|
73 |
|
---|
74 | Args:
|
---|
75 | info (Inform): Contains either a request for action or information.
|
---|
76 | """
|
---|
77 |
|
---|
78 | # a Settings message is the first message that will be send to your
|
---|
79 | # agent containing all the information about the negotiation session.
|
---|
80 | if isinstance(info, Settings):
|
---|
81 | self._settings: Settings = cast(Settings, info)
|
---|
82 | self._me = self._settings.getID()
|
---|
83 |
|
---|
84 | # progress towards the deadline has to be tracked manually through the use of the Progress object
|
---|
85 | self._progress: Progress = self._settings.getProgress()
|
---|
86 |
|
---|
87 | # the profile contains the preferences of the agent over the domain
|
---|
88 | self._profile = ProfileConnectionFactory.create(
|
---|
89 | info.getProfile().getURI(), self.getReporter()
|
---|
90 | )
|
---|
91 |
|
---|
92 | self._bid_list = sorted(AllBidsList(self._profile.getProfile().getDomain()),
|
---|
93 | key=self._profile.getProfile().getUtility, reverse=True)
|
---|
94 | self._opponent_model = freq_opp_mod.FrequencyOpponentModel(self._profile.getProfile().getDomain(), {}, 0,
|
---|
95 | None).With(
|
---|
96 | self._profile.getProfile().getDomain(), None)
|
---|
97 | # ActionDone is an action send by an opponent (an offer or an accept)
|
---|
98 | elif isinstance(info, ActionDone):
|
---|
99 | action: Action = cast(ActionDone, info).getAction()
|
---|
100 |
|
---|
101 | # if it is an offer, set the last received bid
|
---|
102 | if isinstance(action, Offer):
|
---|
103 | bid = cast(Offer, action).getBid()
|
---|
104 | if self._last_sent_bid is None or bid != self._last_sent_bid:
|
---|
105 | self._last_received_bid = bid
|
---|
106 | self._received_bids.append(self._last_received_bid)
|
---|
107 | self._opponent_model = self._opponent_model.WithAction(action, self._progress)
|
---|
108 | # YourTurn notifies you that it is your turn to act
|
---|
109 | elif isinstance(info, YourTurn):
|
---|
110 | action = self._myTurn()
|
---|
111 | if isinstance(self._progress, ProgressRounds):
|
---|
112 | self._progress = self._progress.advance()
|
---|
113 | self.getConnection().send(action)
|
---|
114 |
|
---|
115 | # Finished will be send if the negotiation has ended (through agreement or deadline)
|
---|
116 | elif isinstance(info, Finished):
|
---|
117 | # terminate the agent MUST BE CALLED
|
---|
118 | self.terminate()
|
---|
119 | else:
|
---|
120 | self.getReporter().log(
|
---|
121 | logging.WARNING, "Ignoring unknown info " + str(info)
|
---|
122 | )
|
---|
123 |
|
---|
124 | # lets the geniusweb system know what settings this agent can handle
|
---|
125 | # leave it as it is for this competition
|
---|
126 | def getCapabilities(self) -> Capabilities:
|
---|
127 | return Capabilities(
|
---|
128 | {"SAOP"},
|
---|
129 | {"geniusweb.profile.utilityspace.LinearAdditive"},
|
---|
130 | )
|
---|
131 |
|
---|
132 | # terminates the agent and its connections
|
---|
133 | # leave it as it is for this competition
|
---|
134 | def terminate(self):
|
---|
135 | self.getReporter().log(logging.INFO, "party is terminating:")
|
---|
136 | super().terminate()
|
---|
137 | if self._profile is not None:
|
---|
138 | self._profile.close()
|
---|
139 | self._profile = None
|
---|
140 |
|
---|
141 |
|
---|
142 |
|
---|
143 | # give a description of your agent
|
---|
144 | def getDescription(self) -> str:
|
---|
145 | return """
|
---|
146 | -- Shreker --
|
---|
147 | The Shreker agent is an agent that changes its strategy depending on the time in the following order:
|
---|
148 | - Random walker: initially explores opponent utility space while prevent opponent from getting our best bids
|
---|
149 | - Agreeable: agent by Sahar Mirzayi from ANAC 2018; offers the highest utility bid that concedes on one issue
|
---|
150 | from the offer
|
---|
151 | - Social welfare: late into the negotiation optimizes social welfare if opponent still has not conceded much
|
---|
152 | - Received bids: very late into the negotiation return one of the best bids out of the 20 last received bids"""
|
---|
153 |
|
---|
154 | # execute a turn
|
---|
155 | def _myTurn(self):
|
---|
156 | # Update best received utility
|
---|
157 | if self._last_received_bid is not None and self._best_received_utility < self._profile.getProfile().getUtility(
|
---|
158 | self._last_received_bid):
|
---|
159 | self._best_received_utility = self._profile.getProfile().getUtility(self._last_received_bid)
|
---|
160 | # Find the next bid to send
|
---|
161 | next_sent_bid = self._findBid()
|
---|
162 |
|
---|
163 | # Check whether the bid the bid to be offered follows some specific strategy based on received bids
|
---|
164 | # We do pass a bid we create, it is not an error :)
|
---|
165 | if self._isGood(next_sent_bid):
|
---|
166 | # If the next bid we would send wouldn't improve our chances of getting a better outcome, accept the last
|
---|
167 | # received bid
|
---|
168 | action = Accept(self._me, self._last_received_bid)
|
---|
169 | else:
|
---|
170 | # Otherwise, sent the bid, remove it so we do not send the same bid over and over
|
---|
171 | if next_sent_bid in self._bid_list:
|
---|
172 | self._bid_list.remove(next_sent_bid)
|
---|
173 | self._last_sent_bid = next_sent_bid
|
---|
174 | action = Offer(self._me, next_sent_bid)
|
---|
175 |
|
---|
176 | # send the action
|
---|
177 | return action
|
---|
178 |
|
---|
179 | # Method to check if we want to end the negotiation based on our next bid
|
---|
180 | def _isGood(self, next_sent_bid) -> bool:
|
---|
181 | if len(self._received_bids) == 0:
|
---|
182 | return False
|
---|
183 | profile = self._profile.getProfile()
|
---|
184 |
|
---|
185 | progress = self._progress.get(time.time() * 1000)
|
---|
186 |
|
---|
187 | # Create an acceptance profile and check the metrics used
|
---|
188 | ac = AcceptanceStrategy(progress, profile, self._received_bids, next_sent_bid, self._last_sent_bid)
|
---|
189 | return ac.combi_max_w(self.thresholds[0], 1, 0)
|
---|
190 |
|
---|
191 | # Finds the next bid to send to the opponent
|
---|
192 | # Until threshold[4] -> RandomWalker
|
---|
193 | # threshold[4] until threshold[5] -> AgreeableAgent
|
---|
194 | # threshold[5] until threshold[6] -> SocialWelfareAgent
|
---|
195 | # After threshold[7] -> Send bids we received with best utility
|
---|
196 | def _findBid(self):
|
---|
197 | progress = self._progress.get(time.time() * 1000)
|
---|
198 | profile = self._profile.getProfile()
|
---|
199 | opponent = self._opponent_model
|
---|
200 | # Random Walker above specific threshold
|
---|
201 | if progress < self.thresholds[4]:
|
---|
202 | return self._generateRandomBidAbove(lambda x: x >= self.thresholds[1], self._bid_list, profile.getUtility)
|
---|
203 | # Agreeable agent based on ANAC 2018 agent
|
---|
204 | if progress < self.thresholds[5]:
|
---|
205 | return self._agreeable()
|
---|
206 | # Agent that maximizes the nash product
|
---|
207 | if progress < self.thresholds[6]:
|
---|
208 | return self._socialWelfare(lambda x: (self._profile.getProfile().getUtility(x)) * opponent.getUtility(x))
|
---|
209 | # Send bids that we received and maximize our utility
|
---|
210 | return self._sendReceived()
|
---|
211 |
|
---|
212 | # Function to generate a random bid using a specific thresholding function
|
---|
213 | # threshold_function -> lambda function that returns a boolean used to filter bids
|
---|
214 | # bid_list -> list of bids to chose from
|
---|
215 | # utility_function -> lambda function that computes the utility of a bid
|
---|
216 | def _generateRandomBidAbove(self, threshold_function, bid_list, utility_function):
|
---|
217 | for _ in range(50):
|
---|
218 | bid = self._getRandomBid(bid_list)
|
---|
219 | if threshold_function(utility_function(bid)):
|
---|
220 | return bid
|
---|
221 | return self._bid_list[0]
|
---|
222 |
|
---|
223 | # Generate a random element of the input list
|
---|
224 | def _getRandomBid(self, bid_list) -> Bid:
|
---|
225 | return bid_list[randint(0, len(bid_list) - 1)]
|
---|
226 |
|
---|
227 | # Finds the next bid in the behaviour of the agreeable agent
|
---|
228 | # - gets all bids above a specific time threshold
|
---|
229 | # - selects one of them based on the social welfare (roulette selection)
|
---|
230 | def _agreeable(self) -> Bid:
|
---|
231 | # To collect enough data start by sending the best offers for us
|
---|
232 | target_utility = min(self.thresholds[2], (1 - self._progress.get(time.time() * 1000)) * self.thresholds[3])
|
---|
233 | profile = self._profile.getProfile()
|
---|
234 | bids = []
|
---|
235 | for bid in self._bid_list:
|
---|
236 | if profile.getUtility(bid) > target_utility:
|
---|
237 | bids.append(bid)
|
---|
238 | bids = sorted(bids, key=self._opponent_model.getUtility, reverse=True)
|
---|
239 | if len(bids) == 0:
|
---|
240 | return self._bid_list[0]
|
---|
241 | weights = np.array(
|
---|
242 | [float(profile.getUtility(bid)) + float(self._opponent_model.getUtility(bid)) for bid in bids])
|
---|
243 | return choices(bids, weights=weights / np.sum(weights))[0]
|
---|
244 |
|
---|
245 | # Picks one bid from the bid list that maximizes a specific metric
|
---|
246 | def _socialWelfare(self, metric):
|
---|
247 | best_bid = self._bid_list[0]
|
---|
248 | for bid in self._bid_list:
|
---|
249 | if metric(best_bid) < metric(bid):
|
---|
250 | best_bid = bid
|
---|
251 | return best_bid
|
---|
252 |
|
---|
253 | # Sends bid we have received while maximizing our utility gained from them
|
---|
254 | # Used at the very end to get as much as we can from the negotiation
|
---|
255 | def _sendReceived(self):
|
---|
256 | # Get top 20 received bids and select randomly based on our utility
|
---|
257 | profile = self._profile.getProfile()
|
---|
258 | top_20 = sorted(self._received_bids, key=profile.getUtility, reverse=True)[:20]
|
---|
259 | weights = np.array([float(profile.getUtility(bid)) for bid in top_20])
|
---|
260 | return random.choices(top_20, k=1, weights=weights / np.sum(weights))[0]
|
---|