1 | from decimal import Decimal
|
---|
2 | import logging
|
---|
3 | import json
|
---|
4 | from os import path
|
---|
5 | from random import randint
|
---|
6 | from re import A
|
---|
7 | from time import time
|
---|
8 | from typing import Optional, cast
|
---|
9 |
|
---|
10 | from geniusweb.actions.Accept import Accept
|
---|
11 | from geniusweb.actions.Action import Action
|
---|
12 | from geniusweb.actions.Offer import Offer
|
---|
13 | from geniusweb.actions.PartyId import PartyId
|
---|
14 | from geniusweb.bidspace.AllBidsList import AllBidsList
|
---|
15 | from geniusweb.bidspace.BidsWithUtility import BidsWithUtility
|
---|
16 | from geniusweb.bidspace.Interval import Interval
|
---|
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 numpy import append
|
---|
35 | from tudelft_utilities_logging.ReportToLogger import ReportToLogger
|
---|
36 | from .utils import opponent_model
|
---|
37 |
|
---|
38 | from .utils.opponent_model import OpponentModel
|
---|
39 | from .utils.time_estimator import TimeEstimator
|
---|
40 | from .utils.bid_chooser_2 import BidChooser
|
---|
41 | from .utils.strategy_model import StrategyModel
|
---|
42 |
|
---|
43 | # Some testing flags
|
---|
44 | test_use_accept = True
|
---|
45 | class ProcrastinAgent(DefaultParty):
|
---|
46 | """
|
---|
47 | The Mild Bunch Team's Python geniusweb agent.
|
---|
48 | """
|
---|
49 |
|
---|
50 | def __init__(self):
|
---|
51 | super().__init__()
|
---|
52 | self.logger: ReportToLogger = self.getReporter()
|
---|
53 |
|
---|
54 | self.domain: Domain = None
|
---|
55 | self.parameters: Parameters = None
|
---|
56 | self.profile: LinearAdditiveUtilitySpace = None
|
---|
57 | self.progress: ProgressTime = None
|
---|
58 | self.me: PartyId = None
|
---|
59 | self.other: str = None
|
---|
60 | self.settings: Settings = None
|
---|
61 | self.storage_dir: str = None
|
---|
62 | self.strategy_model = None
|
---|
63 |
|
---|
64 | self.last_received_bid: Bid = None
|
---|
65 | self.opponent_best_bid: Bid = None
|
---|
66 | self.opp_best_self_util: Bid = None
|
---|
67 | self.opponent_concession_bid: Bid = None
|
---|
68 | self.opp_concession_self_util: Bid = 0.0
|
---|
69 | self.alpha: float = 0.5
|
---|
70 | self.lowest_acceptable: float = 1.0
|
---|
71 | self.opponent_model: OpponentModel = None
|
---|
72 | self.bid_chooser: BidChooser = None
|
---|
73 | self.opponent_data: dict = None
|
---|
74 | self.time_estimator: TimeEstimator = TimeEstimator()
|
---|
75 | self.bids_sent: int = 0
|
---|
76 | self.bids_received: int = 0
|
---|
77 | self.logger.log(logging.INFO, "party is initialized")
|
---|
78 |
|
---|
79 | self.test_bids_left = []
|
---|
80 |
|
---|
81 | def getCapabilities(self) -> Capabilities:
|
---|
82 | """MUST BE IMPLEMENTED
|
---|
83 | Method to indicate to the protocol what the capabilities of this agent are.
|
---|
84 | Leave it as is for the ANL 2022 competition
|
---|
85 |
|
---|
86 | Returns:
|
---|
87 | Capabilities: Capabilities representation class
|
---|
88 | """
|
---|
89 | return Capabilities(
|
---|
90 | set(["SAOP"]),
|
---|
91 | set(["geniusweb.profile.utilityspace.LinearAdditive"]),
|
---|
92 | )
|
---|
93 |
|
---|
94 | def getDescription(self) -> str:
|
---|
95 | """MUST BE IMPLEMENTED
|
---|
96 |
|
---|
97 | Returns:
|
---|
98 | str: Agent description
|
---|
99 | """
|
---|
100 | return "The Mild Bunch's ProcrastinAgent for the ANL 2022 competition." \
|
---|
101 | " This agent puts off concesssion till the end of the negotiation."
|
---|
102 | " It's developers are also procrastinagents! The Procrastin-A-Team!"
|
---|
103 |
|
---|
104 | def send_action(self, action: Action):
|
---|
105 | """Sends an action to the opponent(s)
|
---|
106 |
|
---|
107 | Args:
|
---|
108 | action (Action): action of this agent
|
---|
109 | """
|
---|
110 | self.getConnection().send(action)
|
---|
111 |
|
---|
112 | def extract_name(self, party: PartyId) -> str:
|
---|
113 | return str(party).rsplit("_", 1)[0]
|
---|
114 |
|
---|
115 | def current_time(self) -> float:
|
---|
116 | return self.progress.get(time() * 1000)
|
---|
117 |
|
---|
118 | def notifyChange(self, data: Inform):
|
---|
119 | """MUST BE IMPLEMENTED
|
---|
120 | This is the entry point of all interaction with your agent after is has been initialised.
|
---|
121 | How to handle the received data is based on its class type.
|
---|
122 |
|
---|
123 | Args:
|
---|
124 | data (Inform): Contains either a request for action or information.
|
---|
125 | """
|
---|
126 |
|
---|
127 | # a Settings message is the first message that will be send to your
|
---|
128 | # agent containing all the information about the negotiation session.
|
---|
129 | if isinstance(data, Settings):
|
---|
130 | self.settings = cast(Settings, data)
|
---|
131 | self.me = self.settings.getID()
|
---|
132 |
|
---|
133 | # progress towards the deadline has to be tracked manually through the use of the Progress object
|
---|
134 | self.progress = self.settings.getProgress()
|
---|
135 |
|
---|
136 | self.parameters = self.settings.getParameters()
|
---|
137 | self.storage_dir = self.parameters.get("storage_dir")
|
---|
138 |
|
---|
139 | # the profile contains the preferences of the agent over the domain
|
---|
140 | profile_connection = ProfileConnectionFactory.create(
|
---|
141 | data.getProfile().getURI(), self.getReporter()
|
---|
142 | )
|
---|
143 | self.profile = profile_connection.getProfile()
|
---|
144 | self.domain = self.profile.getDomain()
|
---|
145 | profile_connection.close()
|
---|
146 |
|
---|
147 | # ActionDone informs you of an action (an offer or an accept)
|
---|
148 | # that is performed by one of the agents (including yourself).
|
---|
149 | elif isinstance(data, ActionDone):
|
---|
150 | action = cast(ActionDone, data).getAction()
|
---|
151 | actor = action.getActor()
|
---|
152 |
|
---|
153 | # ignore action if it is our action
|
---|
154 | if actor != self.me:
|
---|
155 | if self.other is None:
|
---|
156 | # obtain the name of the opponent, cutting of the position ID.
|
---|
157 | self.other = self.extract_name(actor)
|
---|
158 | # now that the name of the opponent is known, we load our stored data about them
|
---|
159 | self.load_data()
|
---|
160 |
|
---|
161 | # process action done by opponent
|
---|
162 | self.opponent_action(action)
|
---|
163 | # YourTurn notifies you that it is your turn to act
|
---|
164 | elif isinstance(data, YourTurn):
|
---|
165 | # execute a turn
|
---|
166 | self.my_turn()
|
---|
167 |
|
---|
168 | # Finished will be send if the negotiation has ended (through agreement or deadline)
|
---|
169 | elif isinstance(data, Finished):
|
---|
170 | finished = cast(Finished, data)
|
---|
171 | self.save_data(finished)
|
---|
172 | # terminate the agent MUST BE CALLED
|
---|
173 | self.logger.log(logging.INFO, "party is terminating:")
|
---|
174 | super().terminate()
|
---|
175 | else:
|
---|
176 | self.logger.log(logging.WARNING, "Ignoring unknown info " + str(data))
|
---|
177 |
|
---|
178 | def opponent_action(self, action: Action):
|
---|
179 | """Process an action that was received from the opponent.
|
---|
180 |
|
---|
181 | Args:
|
---|
182 | action (Action): action of opponent
|
---|
183 | """
|
---|
184 | if isinstance(action, Accept):
|
---|
185 | # opponent accepted, no response necessary
|
---|
186 | pass
|
---|
187 |
|
---|
188 | # if it is an offer, set the last received bid
|
---|
189 | if isinstance(action, Offer):
|
---|
190 | offer = cast(Offer, action)
|
---|
191 | self.bids_received += 1
|
---|
192 | self.process_opponent_offer(offer)
|
---|
193 |
|
---|
194 | def my_turn(self):
|
---|
195 | """This method is called when it is our turn. It should decide upon an action
|
---|
196 | to perform and send this action to the opponent.
|
---|
197 | """
|
---|
198 | # log opponent time
|
---|
199 | self.time_estimator.self_times_add(self.current_time())
|
---|
200 | # check if the last received offer is good enough
|
---|
201 | if self.choose_accept(self.last_received_bid):
|
---|
202 | # if so, accept the offer
|
---|
203 | action = Accept(self.me, self.last_received_bid)
|
---|
204 | else:
|
---|
205 | # if not, find a bid to propose as counter offer
|
---|
206 | bid = self.choose_bid()
|
---|
207 | action = Offer(self.me, bid)
|
---|
208 | self.bids_sent += 1
|
---|
209 |
|
---|
210 | # log self time
|
---|
211 | self.time_estimator.opp_times_add(self.current_time())
|
---|
212 | # send the action
|
---|
213 | self.send_action(action)
|
---|
214 |
|
---|
215 | ################################################################################################
|
---|
216 | ################################### Our Implementation ###################################
|
---|
217 | ################################################################################################
|
---|
218 |
|
---|
219 | def load_data(self):
|
---|
220 | # load_data is called as soon as the opponent is known.
|
---|
221 | # In the very rare case where the opponent never makes an offer, load_data is never called.
|
---|
222 | if not path.exists(f"{self.storage_dir}/{self.other}.json"):
|
---|
223 | # First round
|
---|
224 | new_data = {}
|
---|
225 | new_data["count"] = 0
|
---|
226 | new_data["self_accepts"] = 0
|
---|
227 | new_data["did_accept"] = []
|
---|
228 | new_data["opponent_accepts"] = 0
|
---|
229 | new_data["no_accepts"] = 0
|
---|
230 | new_data["beta_values"] = []
|
---|
231 | new_data["time_factor"] = 1.0
|
---|
232 | new_data["alphas"] = []
|
---|
233 | new_data["alpha_achieved"] = []
|
---|
234 | self.opponent_data = new_data
|
---|
235 | else:
|
---|
236 | # Not first round
|
---|
237 | with open(f"{self.storage_dir}/{self.other}.json", "r") as f:
|
---|
238 | self.opponent_data = json.load(f)
|
---|
239 | self.time_estimator.update_time_factor(self.opponent_data["time_factor"])
|
---|
240 |
|
---|
241 | def choose_bid(self) -> Bid:
|
---|
242 | if self.bids_sent <= 5:
|
---|
243 | # Action to take before we have a decent estimate at how many turns are left
|
---|
244 | # Send best bid
|
---|
245 | bid_dict = {}
|
---|
246 | for issue, valueset in self.profile.getUtilities().items():
|
---|
247 | bid_dict[issue] = max(self.domain.getValues(issue), key = lambda v: valueset.getUtility(v))
|
---|
248 | best_bid = Bid(bid_dict)
|
---|
249 | return best_bid
|
---|
250 |
|
---|
251 | offers_left = self.time_estimator.turns_left(self.current_time())
|
---|
252 | self.test_bids_left.append(offers_left)
|
---|
253 | return self.bid_chooser.choose_bid(offers_left, self.current_time())
|
---|
254 |
|
---|
255 | def process_opponent_offer(self, offer: Offer):
|
---|
256 | """Process an offer that was received from the opponent.
|
---|
257 |
|
---|
258 | Args:
|
---|
259 | offer (Offer): offer from opponent
|
---|
260 | """
|
---|
261 | # create opponent model if it was not yet initialised
|
---|
262 | if self.opponent_model is None:
|
---|
263 | self.opponent_model = OpponentModel(self.domain)
|
---|
264 | self.bid_chooser = BidChooser(self.profile, self.opponent_model, 0.5) # TODO update lowest acceptable number
|
---|
265 |
|
---|
266 | bid = offer.getBid()
|
---|
267 |
|
---|
268 | # set bid as last received
|
---|
269 | self.last_received_bid = bid
|
---|
270 | # update opponent model with bid
|
---|
271 | self.opponent_model.update(bid, self.current_time())
|
---|
272 |
|
---|
273 | #set opp_highest_bid to either the first or calculated highest (opp POV) opponent bid
|
---|
274 | update_opponent_best = False
|
---|
275 | if self.opponent_best_bid is None:
|
---|
276 | update_opponent_best = True
|
---|
277 | elif self.opponent_model.get_predicted_utility(bid)[0] > self.opponent_model.get_predicted_utility(self.opponent_best_bid)[0]:
|
---|
278 | #update opp_best_bid if new opponent bid is better for them than previous best
|
---|
279 | update_opponent_best = True
|
---|
280 | if update_opponent_best:
|
---|
281 | self.opponent_best_bid = bid
|
---|
282 | self.opp_best_self_util = float(self.profile.getUtility(bid))
|
---|
283 | if self.strategy_model is None:
|
---|
284 | self.strategy_model = StrategyModel(self.opponent_data["alphas"], self.opponent_data["beta_values"], self.opponent_data["did_accept"])
|
---|
285 | self.alpha = self.strategy_model.max_u(self.opp_best_self_util, 0.5, 1.0, mag = 3)
|
---|
286 | self.lowest_acceptable = self.opp_best_self_util + self.alpha * (1.0 - self.opp_best_self_util)
|
---|
287 | self.bid_chooser.update_lowest_acceptable(self.lowest_acceptable)
|
---|
288 |
|
---|
289 | #set opp_concession_bid to either the first or highest (self POV) opponent bid
|
---|
290 | update_opponent_concession = False
|
---|
291 | if self.opponent_concession_bid is None:
|
---|
292 | update_opponent_concession = True
|
---|
293 | elif float(self.profile.getUtility(bid)) > self.opp_concession_self_util:
|
---|
294 | update_opponent_concession = True
|
---|
295 | if update_opponent_concession:
|
---|
296 | self.opponent_concession_bid = bid
|
---|
297 | self.opp_concession_self_util = float(self.profile.getUtility(bid))
|
---|
298 |
|
---|
299 | # update bid_chooser
|
---|
300 | self.bid_chooser.update_bid(bid)
|
---|
301 |
|
---|
302 | def choose_accept(self, bid: Bid) -> bool:
|
---|
303 | if bid is None:
|
---|
304 | return False
|
---|
305 |
|
---|
306 | # very basic approach that accepts if the offer is valued above a certain amount of opponent's highest utility, calculated
|
---|
307 | # through formula t = b + alpha (1.0 - b)
|
---|
308 | # only accepts during last 20 turns or last 1/1000 of the negotiation (or our best was offered)
|
---|
309 | bid_util = float(self.profile.getUtility(bid))
|
---|
310 | time = self.current_time()
|
---|
311 | conditions = [
|
---|
312 | bid_util >= max(self.lowest_acceptable, self.opp_concession_self_util),
|
---|
313 | any([
|
---|
314 | self.time_estimator.turns_left(time) < 20,
|
---|
315 | time > 0.999,
|
---|
316 | bid_util >= 1.0,
|
---|
317 | ]),
|
---|
318 | #test_use_accept, # Tests agent behaviour without accepting
|
---|
319 | ]
|
---|
320 | return all(conditions)
|
---|
321 |
|
---|
322 | def save_data(self, finished: Finished):
|
---|
323 | """This method is called after the negotiation is finished. It can be used to store data
|
---|
324 | for learning capabilities. Note that no extensive calculations can be done within this method.
|
---|
325 | Taking too much time might result in your agent being killed, so use it for storage only.
|
---|
326 | """
|
---|
327 | agreements = list(finished.getAgreements().getAgreements().items())
|
---|
328 | save = self.opponent_data
|
---|
329 | save["count"] += 1
|
---|
330 | save["test_bid_pool_size"] = len(self.bid_chooser.bid_pool)
|
---|
331 | save["test_time_list_self"] = self.time_estimator.self_times
|
---|
332 | #save["test_time_list_opp"] = self.time_estimator.opp_times_adj
|
---|
333 | save["test_offers_left"] = self.test_bids_left
|
---|
334 | save["self_diff"] = self.time_estimator.self_diff
|
---|
335 |
|
---|
336 | opp_stuff = {"weights": {}}
|
---|
337 | total_weight = 0.0
|
---|
338 | for issue in self.domain.getIssues():
|
---|
339 | opp_stuff[issue] = {}
|
---|
340 | total_weight += self.opponent_model.issue_estimators[issue].weight
|
---|
341 | opp_stuff["weights"][issue] = self.opponent_model.issue_estimators[issue].weight
|
---|
342 | for value in self.domain.getValues(issue):
|
---|
343 | opp_stuff[issue][value.getValue()] = self.opponent_model.issue_estimators[issue].get_value_utility(value)
|
---|
344 | for issue in self.domain.getIssues():
|
---|
345 | opp_stuff["weights"][issue] = self.opponent_model.issue_estimators[issue].weight / total_weight
|
---|
346 | save["opponent_model"] = opp_stuff
|
---|
347 |
|
---|
348 | beta = float((self.opp_concession_self_util-self.opp_best_self_util)/(1 - self.opp_best_self_util))
|
---|
349 |
|
---|
350 | save["beta_values"].append(beta)
|
---|
351 |
|
---|
352 | if not agreements:
|
---|
353 | agreement_bid = None
|
---|
354 | agreement_party = None
|
---|
355 | save["time_factor"] = self.time_estimator.get_new_time_factor(self.test_bids_left, len(self.bid_chooser.bid_pool))
|
---|
356 | save["did_accept"].append(False)
|
---|
357 | else:
|
---|
358 | agreement = agreements[0]
|
---|
359 | agreement_bid = agreement[1]
|
---|
360 | agreement_party = agreement[0]
|
---|
361 | save["did_accept"].append(True)
|
---|
362 | if agreement_party is None:
|
---|
363 | # No agreement was made (or rarely they accepted our first bid)
|
---|
364 | save["no_accepts"] += 1
|
---|
365 | elif self.extract_name(agreement_party) == self.extract_name(self.me):
|
---|
366 | # We sent the agreement
|
---|
367 | save["self_accepts"] += 1
|
---|
368 | elif (self.other is not None) and self.extract_name(agreement_party) == self.other:
|
---|
369 | # They accepted
|
---|
370 | save["opponent_accepts"] += 1
|
---|
371 | else:
|
---|
372 | # Only way I can imagine getting here is if we offered
|
---|
373 | # the first bid and the opponent accepted.
|
---|
374 | save["other_accepts"] = save.get("other_accepts", 0) + 1
|
---|
375 | pass
|
---|
376 |
|
---|
377 | if agreement_bid is None:
|
---|
378 | alpha_achieved = 0.0
|
---|
379 | else:
|
---|
380 | alpha_achieved = (float(self.profile.getUtility(agreement_bid)) - self.opp_best_self_util) / (1.0 - self.opp_best_self_util)
|
---|
381 | save["alphas"].append(self.alpha)
|
---|
382 | save["alpha_achieved"].append(alpha_achieved)
|
---|
383 |
|
---|
384 | with open(f"{self.storage_dir}/{self.other}.json", "w") as f:
|
---|
385 | json.dump(save, f) |
---|