1 | import logging
|
---|
2 | import numpy as np
|
---|
3 | from pandas import array
|
---|
4 | from random import randint
|
---|
5 | from sklearn.linear_model import LinearRegression
|
---|
6 | from sklearn.ensemble import RandomForestRegressor
|
---|
7 | from sklearn.ensemble import VotingRegressor
|
---|
8 | from sklearn.neighbors import KNeighborsRegressor
|
---|
9 | from time import time
|
---|
10 | from typing import cast
|
---|
11 | import random
|
---|
12 | from geniusweb.actions.Accept import Accept
|
---|
13 | from geniusweb.actions.Action import Action
|
---|
14 | from geniusweb.actions.Offer import Offer
|
---|
15 | from geniusweb.actions.PartyId import PartyId
|
---|
16 | from geniusweb.bidspace.AllBidsList import AllBidsList
|
---|
17 | from geniusweb.inform.ActionDone import ActionDone
|
---|
18 | from geniusweb.inform.Finished import Finished
|
---|
19 | from geniusweb.inform.Inform import Inform
|
---|
20 | from geniusweb.inform.Settings import Settings
|
---|
21 | from geniusweb.inform.YourTurn import YourTurn
|
---|
22 | from geniusweb.issuevalue.Bid import Bid
|
---|
23 | from geniusweb.issuevalue.Domain import Domain
|
---|
24 | from geniusweb.party.Capabilities import Capabilities
|
---|
25 | from geniusweb.party.DefaultParty import DefaultParty
|
---|
26 | from geniusweb.profile.utilityspace.LinearAdditiveUtilitySpace import (
|
---|
27 | LinearAdditiveUtilitySpace,
|
---|
28 | )
|
---|
29 | from geniusweb.profileconnection.ProfileConnectionFactory import (
|
---|
30 | ProfileConnectionFactory,
|
---|
31 | )
|
---|
32 | from geniusweb.progress.ProgressTime import ProgressTime
|
---|
33 | from geniusweb.references.Parameters import Parameters
|
---|
34 | from tudelft_utilities_logging.ReportToLogger import ReportToLogger
|
---|
35 |
|
---|
36 | from agents.template_agent.utils.opponent_model import OpponentModel
|
---|
37 |
|
---|
38 |
|
---|
39 | class BIU_agent(DefaultParty):
|
---|
40 | """
|
---|
41 | BIU_agent of a Python geniusweb agent.
|
---|
42 | """
|
---|
43 | def __init__(self):
|
---|
44 | super().__init__()
|
---|
45 | self.logger: ReportToLogger = self.getReporter()
|
---|
46 |
|
---|
47 | self.domain: Domain = None
|
---|
48 | self.parameters: Parameters = None
|
---|
49 | self.profile: LinearAdditiveUtilitySpace = None
|
---|
50 | self.progress: ProgressTime = None
|
---|
51 | self.me: PartyId = None
|
---|
52 | self.other: str = None
|
---|
53 | self.settings: Settings = None
|
---|
54 | self.storage_dir: str = None
|
---|
55 |
|
---|
56 | self.last_received_bid: Bid = None
|
---|
57 | self.opponent_model: OpponentModel = None
|
---|
58 | self.logger.log(logging.INFO, "party is initialized")
|
---|
59 |
|
---|
60 | self.bids_given: list = None
|
---|
61 | self.bids_received: list = None
|
---|
62 | self.proposal_time: float = None
|
---|
63 | self.opponent_bid_times: list = None
|
---|
64 |
|
---|
65 | def notifyChange(self, data: Inform):
|
---|
66 | """MUST BE IMPLEMENTED
|
---|
67 | This is the entry point of all interaction with your agent after is has been initialised.
|
---|
68 | How to handle the received data is based on its class type.
|
---|
69 |
|
---|
70 | Args:
|
---|
71 | info (Inform): Contains either a request for action or information.
|
---|
72 | """
|
---|
73 |
|
---|
74 | # a Settings message is the first message that will be send to your
|
---|
75 | # agent containing all the information about the negotiation session.
|
---|
76 | if isinstance(data, Settings):
|
---|
77 | self.settings = cast(Settings, data)
|
---|
78 | self.me = self.settings.getID()
|
---|
79 |
|
---|
80 | # progress towards the deadline has to be tracked manually through the use of the Progress object
|
---|
81 | self.progress = self.settings.getProgress()
|
---|
82 |
|
---|
83 | self.parameters = self.settings.getParameters()
|
---|
84 | self.storage_dir = self.parameters.get("storage_dir")
|
---|
85 |
|
---|
86 | # the profile contains the preferences of the agent over the domain
|
---|
87 | profile_connection = ProfileConnectionFactory.create(
|
---|
88 | data.getProfile().getURI(), self.getReporter()
|
---|
89 | )
|
---|
90 | self.profile = profile_connection.getProfile()
|
---|
91 | self.domain = self.profile.getDomain()
|
---|
92 | profile_connection.close()
|
---|
93 |
|
---|
94 | self.opponent_bid_times = []
|
---|
95 |
|
---|
96 | # ActionDone informs you of an action (an offer or an accept)
|
---|
97 | # that is performed by one of the agents (including yourself).
|
---|
98 | elif isinstance(data, ActionDone):
|
---|
99 | action = cast(ActionDone, data).getAction()
|
---|
100 | actor = action.getActor()
|
---|
101 |
|
---|
102 | # ignore action if it is our action
|
---|
103 | if actor != self.me:
|
---|
104 | # obtain the name of the opponent, cutting of the position ID.
|
---|
105 | self.other = str(actor).rsplit("_", 1)[0]
|
---|
106 |
|
---|
107 | # process action done by opponent
|
---|
108 | self.opponent_action(action)
|
---|
109 | # YourTurn notifies you that it is your turn to act
|
---|
110 | elif isinstance(data, YourTurn):
|
---|
111 | # execute a turn
|
---|
112 | if self.proposal_time is not None:
|
---|
113 | self.opponent_bid_times.append(self.progress.get(time() * 1000) - self.proposal_time)
|
---|
114 | self.my_turn()
|
---|
115 | self.proposal_time = self.progress.get(time() * 1000)
|
---|
116 |
|
---|
117 | # Finished will be send if the negotiation has ended (through agreement or deadline)
|
---|
118 | elif isinstance(data, Finished):
|
---|
119 | self.save_data()
|
---|
120 | # terminate the agent MUST BE CALLED
|
---|
121 | self.logger.log(logging.INFO, "party is terminating:")
|
---|
122 | super().terminate()
|
---|
123 | else:
|
---|
124 | self.logger.log(logging.WARNING, "Ignoring unknown info " + str(data))
|
---|
125 |
|
---|
126 | def getCapabilities(self) -> Capabilities:
|
---|
127 | """MUST BE IMPLEMENTED
|
---|
128 | Method to indicate to the protocol what the capabilities of this agent are.
|
---|
129 | Leave it as is for the ANL 2022 competition
|
---|
130 |
|
---|
131 | Returns:
|
---|
132 | Capabilities: Capabilities representation class
|
---|
133 | """
|
---|
134 | return Capabilities(
|
---|
135 | set(["SAOP"]),
|
---|
136 | set(["geniusweb.profile.utilityspace.LinearAdditive"]),
|
---|
137 | )
|
---|
138 |
|
---|
139 | def send_action(self, action: Action):
|
---|
140 | """Sends an action to the opponent(s)
|
---|
141 |
|
---|
142 | Args:
|
---|
143 | action (Action): action of this agent
|
---|
144 | """
|
---|
145 | self.getConnection().send(action)
|
---|
146 |
|
---|
147 | # give a description of your agent
|
---|
148 | def getDescription(self) -> str:
|
---|
149 | """MUST BE IMPLEMENTED
|
---|
150 | Returns a description of your agent. 1 or 2 sentences.
|
---|
151 |
|
---|
152 | Returns:
|
---|
153 | str: Agent description
|
---|
154 | """
|
---|
155 | return "This is a Bar Ilan University agent that learns from the opponent's bids, by using a random forest, a linear regression and a KNN. The agent also using random stochastic to take the offers."
|
---|
156 |
|
---|
157 | def opponent_action(self, action):
|
---|
158 | """Process an action that was received from the opponent.
|
---|
159 |
|
---|
160 | Args:
|
---|
161 | action (Action): action of opponent
|
---|
162 | """
|
---|
163 | # if it is an offer, set the last received bid
|
---|
164 | if isinstance(action, Offer):
|
---|
165 | # create opponent model if it was not yet initialised
|
---|
166 | if self.opponent_model is None:
|
---|
167 | self.opponent_model = OpponentModel(self.domain)
|
---|
168 |
|
---|
169 | bid = cast(Offer, action).getBid()
|
---|
170 |
|
---|
171 | # update opponent model with bid
|
---|
172 | self.opponent_model.update(bid)
|
---|
173 | # set bid as last received
|
---|
174 | self.last_received_bid = bid
|
---|
175 |
|
---|
176 | def my_turn(self):
|
---|
177 | """This method is called when it is our turn. It should decide upon an action
|
---|
178 | to perform and send this action to the opponent.
|
---|
179 | """
|
---|
180 | if self.accept_condition(self.last_received_bid):
|
---|
181 | action = Accept(self.me, self.last_received_bid)
|
---|
182 | else:
|
---|
183 | t = self.progress.get(time() * 1000)
|
---|
184 | self.logger.log(logging.INFO, t)
|
---|
185 | bid = self.find_bid()
|
---|
186 | if t >= 0.95:
|
---|
187 | t_o = self.regression_opponent_time(self.opponent_bid_times[-10:])
|
---|
188 | self.logger.log(logging.INFO, self.opponent_bid_times)
|
---|
189 | self.logger.log(logging.INFO, t_o)
|
---|
190 | while all(t < 1 - t_o):
|
---|
191 | t = self.progress.get(time() * 1000)
|
---|
192 | action = Offer(self.me, bid)
|
---|
193 |
|
---|
194 | self.send_action(action)
|
---|
195 |
|
---|
196 | def save_data(self):
|
---|
197 | """This method is called after the negotiation is finished. It can be used to store data
|
---|
198 | for learning capabilities. Note that no extensive calculations can be done within this method.
|
---|
199 | Taking too much time might result in your agent being killed, so use it for storage only.
|
---|
200 | """
|
---|
201 | data = " ".join(str(x) for x in self.opponent_bid_times)
|
---|
202 | # self_dir = "./agents/BIU_agent/data.md"
|
---|
203 | with open(f"{self.storage_dir}/data.md", "w") as f:
|
---|
204 | f.write(data)
|
---|
205 |
|
---|
206 | ###########################################################################################
|
---|
207 | ################################## Example methods below ##################################
|
---|
208 | ###########################################################################################
|
---|
209 |
|
---|
210 | def accept_condition(self, bid: Bid) -> bool:
|
---|
211 | if bid is None:
|
---|
212 | return False
|
---|
213 |
|
---|
214 | # progress of the negotiation session between 0 and 1 (1 is deadline)
|
---|
215 | progress = self.progress.get(time() * 1000)
|
---|
216 |
|
---|
217 | # very basic approach that accepts if the offer is valued above 0.7 and
|
---|
218 | # 95% of the time towards the deadline has passed
|
---|
219 | threshold = 0.9
|
---|
220 | if 0 < progress < 0.2:
|
---|
221 | threshold = 0.9
|
---|
222 | if 0.2 < progress <0.3:
|
---|
223 | threshold = 0.8
|
---|
224 | elif 0.3 < progress < 0.5:
|
---|
225 | threshold = 0.6
|
---|
226 | elif 0.5 < progress < 0.9:
|
---|
227 | threshold = 0.5
|
---|
228 | elif 0.9 < progress < 1:
|
---|
229 | threshold = 0.25
|
---|
230 |
|
---|
231 | conditions = [
|
---|
232 | self.profile.getUtility(bid) > 0.8
|
---|
233 | ]
|
---|
234 | return all(conditions)
|
---|
235 |
|
---|
236 | def find_bid(self) -> Bid:
|
---|
237 | # compose a list of all possible bids
|
---|
238 | domain = self.profile.getDomain()
|
---|
239 | all_bids = AllBidsList(domain)
|
---|
240 |
|
---|
241 | best_bid_score = 0.0
|
---|
242 | best_bid = None
|
---|
243 |
|
---|
244 | # take 500 attempts to find a bid according to a heuristic score
|
---|
245 | for _ in range(500):
|
---|
246 | bid = all_bids.get(randint(0, all_bids.size() - 1))
|
---|
247 | bid_score = self.score_bid(bid)
|
---|
248 | if bid_score > best_bid_score:
|
---|
249 | best_bid_score, best_bid = bid_score, bid
|
---|
250 |
|
---|
251 | return best_bid
|
---|
252 |
|
---|
253 | def score_bid(self, bid: Bid, alpha: float = 0.95, eps: float = 0.5) -> float:
|
---|
254 | """Calculate heuristic score for a bid
|
---|
255 |
|
---|
256 | Args:
|
---|
257 | bid (Bid): Bid to score
|
---|
258 | alpha (float, optional): Trade-off factor between self interested and
|
---|
259 | altruistic behaviour. Defaults to 0.95.
|
---|
260 | eps (float, optional): Time pressure factor, balances between conceding
|
---|
261 | and Boulware behaviour over time. Defaults to 0.1.
|
---|
262 |
|
---|
263 | Returns:
|
---|
264 | float: score
|
---|
265 | """
|
---|
266 |
|
---|
267 | # progress = self.progress.get(time() * 1000)
|
---|
268 |
|
---|
269 | # our_utility = float(self.profile.getUtility(bid))
|
---|
270 |
|
---|
271 | # time_pressure = 1.0 - progress ** (1 / eps)
|
---|
272 | # score = alpha * time_pressure * our_utility
|
---|
273 |
|
---|
274 | # if self.opponent_model is not None:
|
---|
275 | # opponent_utility = self.opponent_model.get_predicted_utility(bid)
|
---|
276 | # opponent_score = (1.0 - alpha * time_pressure) * opponent_utility
|
---|
277 | # score += opponent_score
|
---|
278 |
|
---|
279 | # return our_utility
|
---|
280 | stochastic_alpha = 0
|
---|
281 | stochastic_eps = 0
|
---|
282 | STOCHASTIC_TRANSITION = random.randint(0,9)
|
---|
283 | if 0 < STOCHASTIC_TRANSITION < 9: # alpha stay the same
|
---|
284 | stochastic_alpha = alpha
|
---|
285 | elif STOCHASTIC_TRANSITION == 0:
|
---|
286 | stochastic_alpha = alpha - eps
|
---|
287 | stochastic_eps = 0.005
|
---|
288 | else: # STOCHASTIC_TRANSITION = 9
|
---|
289 | stochastic_alpha = alpha + eps
|
---|
290 | stochastic_eps = -0.005
|
---|
291 |
|
---|
292 | progress = self.progress.get(time() * 1000)
|
---|
293 |
|
---|
294 | utility = float(self.profile.getUtility(bid))
|
---|
295 |
|
---|
296 | time_pressure = 1.0 - progress ** (1 / eps)
|
---|
297 | score = stochastic_alpha * time_pressure * utility
|
---|
298 |
|
---|
299 | if self.opponent_model is not None:
|
---|
300 | opponent_utility = self.opponent_model.get_predicted_utility(bid)
|
---|
301 | opponent_score = (1.0 - stochastic_alpha * time_pressure) * opponent_utility
|
---|
302 | score += opponent_score
|
---|
303 | if utility > 0.994 and stochastic_eps > 0:
|
---|
304 | stochastic_eps = 0
|
---|
305 | if utility < 0.005 and stochastic_eps < 0:
|
---|
306 | stochastic_eps = 0
|
---|
307 | final_score = utility + stochastic_eps
|
---|
308 | return final_score
|
---|
309 |
|
---|
310 |
|
---|
311 |
|
---|
312 | def regression_opponent_time(self, bid_times):
|
---|
313 | r1 = LinearRegression()
|
---|
314 | r2 = RandomForestRegressor(n_estimators=10, random_state=1)
|
---|
315 | r3 = KNeighborsRegressor()
|
---|
316 | X = array(range(len(bid_times))).reshape(-1, 1)
|
---|
317 | y = array(bid_times).reshape(-1, 1)
|
---|
318 | er = VotingRegressor([('lr', r1), ('rf', r2), ('r3', r3)])
|
---|
319 | return er.fit(X, y).predict(X) |
---|