Spaces:
Runtime error
Runtime error
File size: 42,052 Bytes
8acb22e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 |
from typing import List
from langchain.base_language import BaseLanguageModel
from langchain.schema import (
AIMessage,
HumanMessage,
SystemMessage
)
from langchain.chat_models import (
ChatAnthropic,
ChatOpenAI,
ChatVertexAI,
ChatGooglePalm,
)
import vertexai
from langchain.input import get_colored_text
from langchain.callbacks import get_openai_callback
from collections import defaultdict
from pydantic import BaseModel
import queue
import threading
import os
import random
import time
import ujson as json
import matplotlib.pyplot as plt
from .item_base import Item, item_list_equal
from .prompt_base import (
AUCTION_HISTORY,
# INSTRUCT_OBSERVE_TEMPLATE,
_LEARNING_STATEMENT,
INSTRUCT_PLAN_TEMPLATE,
INSTRUCT_BID_TEMPLATE,
INSTRUCT_SUMMARIZE_TEMPLATE,
INSTRUCT_LEARNING_TEMPLATE,
INSTRUCT_REPLAN_TEMPLATE,
SYSTEM_MESSAGE,
)
import sys
sys.path.append('..')
from utils import LoadJsonL, extract_jsons_from_text, extract_numbered_list, trace_back
# DESIRE_DESC = {
# 'default': "Your goal is to fully utilize your budget while actively participating in the auction",
# 'maximize_profit': "Your goal is to maximize your overall profit, and fully utilize your budget while actively participating in the auction. This involves strategic bidding to win items for less than their true value, thereby ensuring the difference between the price paid and the item's value is as large as possible",
# 'maximize_items': "Your goal is to win as many items as possible, and fully utilize your budget while actively participating in the auction. While keeping your budget in mind, you should aim to participate broadly across different items, striving to be the highest bidder more often than not",
# } # remove period at the end of each description
DESIRE_DESC = {
'maximize_profit': "Your primary objective is to secure the highest profit at the end of this auction, compared to all other bidders",
'maximize_items': "Your primary objective is to win the highest number of items at the end of this auction, compared to everyone else",
}
class Bidder(BaseModel):
name: str
model_name: str
budget: int
desire: str
plan_strategy: str
temperature: float = 0.7
overestimate_percent: int = 10
correct_belief: bool
enable_learning: bool = False
llm: BaseLanguageModel = None
openai_cost = 0
llm_token_count = 0
verbose: bool = False
auction_hash: str = ''
system_message: str = ''
original_budget: int = 0
# working memory
profit: int = 0
cur_item_id = 0
items: list = []
dialogue_history: list = [] # for gradio UI display
llm_prompt_history: list = [] # for tracking llm calling
items_won = []
bid_history: list = [] # history of the bidding of a single item
plan_instruct: str = '' # instruction for planning
cur_plan: str = '' # current plan
status_quo: dict = {} # belief of budget and profit, self and others
withdraw: bool = False # state of withdraw
learnings: str = '' # learnings from previous biddings. If given, then use it to guide the rest of the auction.
max_bid_cnt: int = 4 # Rule Bidder: maximum number of bids on one item (K = 1 starting bid + K-1 increase bid)
rule_bid_cnt: int = 0 # Rule Bidder: count of bids on one item
# belief tracking
failed_bid_cnt: int = 0 # count of failed bids (overspending)
total_bid_cnt: int = 0 # count of total bids
self_belief_error_cnt: int = 0
total_self_belief_cnt: int = 0
other_belief_error_cnt: int = 0
total_other_belief_cnt: int = 0
engagement_count: int = 0
budget_history = []
profit_history = []
budget_error_history = []
profit_error_history = []
win_bid_error_history = []
engagement_history = defaultdict(int)
all_bidders_status = {} # track others' profit
changes_of_plan = []
# not used
input_box: str = None
need_input = False
semaphore = 0
class Config:
arbitrary_types_allowed = True
def __repr__(self):
return self.name
def __str__(self):
return self.name
@classmethod
def create(cls, **data):
instance = cls(**data)
instance._post_init()
return instance
def _post_init(self):
self.original_budget = self.budget
self.system_message = SYSTEM_MESSAGE.format(
name=self.name,
desire_desc=DESIRE_DESC[self.desire],
)
self._parse_llm()
self.dialogue_history += [
SystemMessage(content=self.system_message),
AIMessage(content='')
]
self.budget_history.append(self.budget)
self.profit_history.append(self.profit)
def _parse_llm(self):
if 'gpt-' in self.model_name:
self.llm = ChatOpenAI(model=self.model_name, temperature=self.temperature, max_retries=30, request_timeout=1200)
elif 'claude' in self.model_name:
self.llm = ChatAnthropic(model=self.model_name, temperature=self.temperature, default_request_timeout=1200)
elif 'bison' in self.model_name:
self.llm = ChatGooglePalm(model_name=f'models/{self.model_name}', temperature=self.temperature)
elif 'rule' in self.model_name or 'human' in self.model_name:
self.llm = None
else:
raise NotImplementedError(self.model_name)
# def _rotate_openai_org(self):
# # use two organizations to avoid rate limit
# if os.environ.get('OPENAI_ORGANIZATION_1') and os.environ.get('OPENAI_ORGANIZATION_2'):
# return random.choice([os.environ.get('OPENAI_ORGANIZATION_1'), os.environ.get('OPENAI_ORGANIZATION_2')])
# else:
# return None
def _run_llm_standalone(self, messages: list):
with get_openai_callback() as cb:
for i in range(6):
try:
input_token_num = self.llm.get_num_tokens_from_messages(messages)
if 'claude' in self.model_name: # anthropic's claude
result = self.llm(messages, max_tokens_to_sample=2048)
elif 'bison' in self.model_name: # google's palm-2
max_tokens = min(max(3900 - input_token_num, 192), 2048)
if isinstance(self.llm, ChatVertexAI):
result = self.llm(messages, max_output_tokens=max_tokens)
else:
result = self.llm(messages)
elif 'gpt' in self.model_name: # openai
if 'gpt-3.5-turbo' in self.model_name and '16k' not in self.model_name:
max_tokens = max(3900 - input_token_num, 192)
else:
# gpt-4
# self.llm.openai_organization = self._rotate_openai_org()
max_tokens = max(8000 - input_token_num, 192)
result = self.llm(messages, max_tokens=max_tokens)
elif 'llama' in self.model_name.lower():
raise NotImplementedError
else:
raise NotImplementedError
break
except:
print(f'Retrying for {self.model_name} ({i+1}/6), wait for {2**(i+1)} sec...')
time.sleep(2**(i+1))
self.openai_cost += cb.total_cost
self.llm_token_count = self.llm.get_num_tokens_from_messages(messages)
return result.content
def _get_estimated_value(self, item):
value = item.true_value * (1 + self.overestimate_percent / 100)
return int(value)
def _get_cur_item(self, key=None):
if self.cur_item_id < len(self.items):
if key is not None:
return self.items[self.cur_item_id].__dict__[key]
else:
return self.items[self.cur_item_id]
else:
return 'no item left'
def _get_next_item(self, key=None):
if self.cur_item_id + 1 < len(self.items):
if key is not None:
return self.items[self.cur_item_id + 1].__dict__[key]
else:
return self.items[self.cur_item_id + 1]
else:
return 'no item left'
def _get_remaining_items(self, as_str=False):
remain_items = self.items[self.cur_item_id + 1:]
if as_str:
return ', '.join([item.name for item in remain_items])
else:
return remain_items
def _get_items_value_str(self, items: List[Item]):
if not isinstance(items, list):
items = [items]
items_info = ''
for i, item in enumerate(items):
estimated_value = self._get_estimated_value(item)
_info = f"{i+1}. {item}, starting price is ${item.price}. Your estimated value for this item is ${estimated_value}.\n"
items_info += _info
return items_info.strip()
# ********** Main Instructions and Functions ********** #
def learn_from_prev_auction(self, past_learnings, past_auction_log):
if not self.enable_learning or 'rule' in self.model_name or 'human' in self.model_name:
return ''
instruct_learn = INSTRUCT_LEARNING_TEMPLATE.format(
past_auction_log=past_auction_log,
past_learnings=past_learnings)
result = self._run_llm_standalone([HumanMessage(content=instruct_learn)])
self.dialogue_history += [
HumanMessage(content=instruct_learn),
AIMessage(content=result),
]
self.llm_prompt_history.append({
'messages': [{x.type: x.content} for x in [HumanMessage(content=instruct_learn)]],
'result': result,
'tag': 'learn_0'
})
self.learnings = '\n'.join(extract_numbered_list(result))
if self.learnings != '':
self.system_message += f"\n\nHere are your key learning points and practical tips from a previous auction. You can use them to guide this auction:\n```\n{self.learnings}\n```"
if self.verbose:
print(f"Learn from previous auction: {self.name} ({self.model_name}).")
return result
def _choose_items(self, budget, items: List[Item]):
'''
Choose items within budget for rule bidders.
Cheap ones first if maximize_items, expensive ones first if maximize_profit.
'''
sorted_items = sorted(items, key=lambda x: self._get_estimated_value(x),
reverse=self.desire == 'maximize_profit')
chosen_items = []
i = 0
while budget >= 0 and i < len(sorted_items):
item = sorted_items[i]
if item.price <= budget:
chosen_items.append(item)
budget -= item.price
i += 1
return chosen_items
def get_plan_instruct(self, items: List[Item]):
self.items = items
plan_instruct = INSTRUCT_PLAN_TEMPLATE.format(
bidder_name=self.name,
budget=self.budget,
item_num=len(items),
items_info=self._get_items_value_str(items),
desire_desc=DESIRE_DESC[self.desire],
learning_statement='' if not self.enable_learning else _LEARNING_STATEMENT
)
return plan_instruct
def init_plan(self, plan_instruct: str):
'''
Plan for bidding with auctioneer's instruction and items information for customize estimated value.
plan = plan(system_message, instruct_plan)
'''
if 'rule' in self.model_name:
# self.cur_plan = ', '.join([x.name for x in self._choose_items(self.budget, self.items)])
# self.dialogue_history += [
# HumanMessage(content=plan_instruct),
# AIMessage(content=self.cur_plan),
# ]
# return self.cur_plan
return ''
self.status_quo = {
'remaining_budget': self.budget,
'total_profits': {bidder: 0 for bidder in self.all_bidders_status.keys()},
'winning_bids': {bidder: {} for bidder in self.all_bidders_status.keys()},
}
if self.plan_strategy == 'none':
self.plan_instruct = ''
self.cur_plan = ''
return None
system_msg = SystemMessage(content=self.system_message)
plan_msg = HumanMessage(content=plan_instruct)
messages = [system_msg, plan_msg]
result = self._run_llm_standalone(messages)
if self.verbose:
print(get_colored_text(plan_msg.content, 'red'))
print(get_colored_text(result, 'green'))
self.dialogue_history += [
plan_msg,
AIMessage(content=result),
]
self.llm_prompt_history.append({
'messages': [{x.type: x.content} for x in messages],
'result': result,
'tag': 'plan_0'
})
self.cur_plan = result
self.plan_instruct = plan_instruct
self.changes_of_plan.append([
f"{self.cur_item_id} (Initial)",
False,
json.dumps(extract_jsons_from_text(result)[-1]),
])
if self.verbose:
print(f"Plan: {self.name} ({self.model_name}) for {self._get_cur_item()}.")
return result
def get_rebid_instruct(self, auctioneer_msg: str):
self.dialogue_history += [
HumanMessage(content=auctioneer_msg),
AIMessage(content='')
]
return auctioneer_msg
def get_bid_instruct(self, auctioneer_msg: str, bid_round: int):
auctioneer_msg = auctioneer_msg.replace(self.name, f'You ({self.name})')
bid_instruct = INSTRUCT_BID_TEMPLATE.format(
auctioneer_msg=auctioneer_msg,
bidder_name=self.name,
cur_item=self._get_cur_item(),
estimated_value=self._get_estimated_value(self._get_cur_item()),
desire_desc=DESIRE_DESC[self.desire],
learning_statement='' if not self.enable_learning else _LEARNING_STATEMENT
)
if bid_round == 0:
if self.plan_strategy in ['static', 'none']:
# if static planner, then no replanning is needed. status quo is updated in replanning. thus need to add status quo in bid instruct.
bid_instruct = f"""The status quo of this auction so far is:\n"{json.dumps(self.status_quo, indent=4)}"\n\n{bid_instruct}\n---\n"""
else:
bid_instruct = f'Now, the auctioneer says: "{auctioneer_msg}"'
self.dialogue_history += [
HumanMessage(content=bid_instruct),
AIMessage(content='')
]
return bid_instruct
def bid_rule(self, cur_bid: int, min_markup_pct: float = 0.1):
'''
:param cur_bid: current highest bid
:param min_markup_pct: minimum percentage for bid increase
:param max_bid_cnt: maximum number of bids on one item (K = 1 starting bid + K-1 increase bid)
'''
# dialogue history already got bid_instruction.
cur_item = self._get_cur_item()
if cur_bid <= 0:
next_bid = cur_item.price
else:
next_bid = cur_bid + min_markup_pct * cur_item.price
if self.budget - next_bid >= 0 and self.rule_bid_cnt < self.max_bid_cnt:
msg = int(next_bid)
self.rule_bid_cnt += 1
else:
msg = -1
content = f'The current highest bid for {cur_item.name} is ${cur_bid}. '
content += "I'm out!" if msg < 0 else f"I bid ${msg}! (Rule generated)"
self.dialogue_history += [
HumanMessage(content=''),
AIMessage(content=content)
]
return msg
def bid(self, bid_instruct):
'''
Bid for an item with auctioneer's instruction and bidding history.
bid_history = bid(system_message, instruct_plan, plan, bid_history)
'''
if self.model_name == 'rule':
return ''
bid_msg = HumanMessage(content=bid_instruct)
if self.plan_strategy == 'none':
messages = [SystemMessage(content=self.system_message)]
else:
messages = [SystemMessage(content=self.system_message),
HumanMessage(content=self.plan_instruct),
AIMessage(content=self.cur_plan)]
self.bid_history += [bid_msg]
messages += self.bid_history
result = self._run_llm_standalone(messages)
self.bid_history += [AIMessage(content=result)]
self.dialogue_history += [
HumanMessage(content=''),
AIMessage(content=result)
]
self.llm_prompt_history.append({
'messages': [{x.type: x.content} for x in messages],
'result': result,
'tag': f'bid_{self.cur_item_id}'
})
if self.verbose:
print(get_colored_text(bid_instruct, 'yellow'))
print(get_colored_text(result, 'green'))
print(f"Bid: {self.name} ({self.model_name}) for {self._get_cur_item()}.")
self.total_bid_cnt += 1
return result
def get_summarize_instruct(self, bidding_history: str, hammer_msg: str, win_lose_msg: str):
instruct = INSTRUCT_SUMMARIZE_TEMPLATE.format(
cur_item=self._get_cur_item(),
bidding_history=bidding_history,
hammer_msg=hammer_msg.strip(),
win_lose_msg=win_lose_msg.strip(),
bidder_name=self.name,
prev_status=self._status_json_to_text(self.status_quo),
)
return instruct
def summarize(self, instruct_summarize: str):
'''
Update belief/status quo
status_quo = summarize(system_message, bid_history, prev_status + instruct_summarize)
'''
self.budget_history.append(self.budget)
self.profit_history.append(self.profit)
if self.model_name == 'rule':
self.rule_bid_cnt = 0 # reset bid count for rule bidder
return ''
messages = [SystemMessage(content=self.system_message)]
# messages += self.bid_history
summ_msg = HumanMessage(content=instruct_summarize)
messages.append(summ_msg)
status_quo_text = self._run_llm_standalone(messages)
self.dialogue_history += [summ_msg, AIMessage(content=status_quo_text)]
self.bid_history += [summ_msg, AIMessage(content=status_quo_text)]
self.llm_prompt_history.append({
'messages': [{x.type: x.content} for x in messages],
'result': status_quo_text,
'tag': f'summarize_{self.cur_item_id}'
})
cnt = 0
while cnt <= 3:
sanity_msg = self._sanity_check_status_json(extract_jsons_from_text(status_quo_text)[-1])
if sanity_msg == '':
# pass sanity check then track beliefs
consistency_msg = self._belief_tracking(status_quo_text)
else:
sanity_msg = f'- {sanity_msg}'
consistency_msg = ''
if sanity_msg != '' or (consistency_msg != '' and self.correct_belief):
err_msg = f"As {self.name}, here are some error(s) of your summary of the status JSON:\n{sanity_msg.strip()}\n{consistency_msg.strip()}\n\nPlease revise the status JSON based on the errors. Don't apologize. Just give me the revised status JSON.".strip()
# print(f"{self.name}: revising status quo for the {cnt} time:")
# print(get_colored_text(err_msg, 'green'))
# print(get_colored_text(status_quo_text, 'red'))
messages += [AIMessage(content=status_quo_text),
HumanMessage(content=err_msg)]
status_quo_text = self._run_llm_standalone(messages)
self.dialogue_history += [
HumanMessage(content=err_msg),
AIMessage(content=status_quo_text),
]
cnt += 1
else:
break
self.status_quo = extract_jsons_from_text(status_quo_text)[-1]
if self.verbose:
print(get_colored_text(instruct_summarize, 'blue'))
print(get_colored_text(status_quo_text, 'green'))
print(f"Summarize: {self.name} ({self.model_name}) for {self._get_cur_item()}.")
return status_quo_text
def get_replan_instruct(self):
instruct = INSTRUCT_REPLAN_TEMPLATE.format(
status_quo=self._status_json_to_text(self.status_quo),
remaining_items_info=self._get_items_value_str(self._get_remaining_items()),
bidder_name=self.name,
desire_desc=DESIRE_DESC[self.desire],
learning_statement='' if not self.enable_learning else _LEARNING_STATEMENT
)
return instruct
def replan(self, instruct_replan: str):
'''
plan = replan(system_message, instruct_plan, prev_plan, status_quo + (learning) + instruct_replan)
'''
if self.model_name == 'rule':
self.withdraw = False
self.cur_item_id += 1
return ''
if self.plan_strategy in ['none', 'static']:
self.bid_history = [] # clear bid history
self.cur_item_id += 1
self.withdraw = False
return 'Skip replanning for bidders with static or no plan.'
replan_msg = HumanMessage(content=instruct_replan)
messages = [SystemMessage(content=self.system_message),
HumanMessage(content=self.plan_instruct),
AIMessage(content=self.cur_plan)]
messages.append(replan_msg)
result = self._run_llm_standalone(messages)
new_plan_dict = extract_jsons_from_text(result)[-1]
cnt = 0
while len(new_plan_dict) == 0 and cnt < 2:
err_msg = 'Your response does not contain a JSON-format priority list for items. Please revise your plan.'
messages += [
AIMessage(content=result),
HumanMessage(content=err_msg),
]
result = self._run_llm_standalone(messages)
new_plan_dict = extract_jsons_from_text(result)[-1]
self.dialogue_history += [
HumanMessage(content=err_msg),
AIMessage(content=result),
]
cnt += 1
old_plan_dict = extract_jsons_from_text(self.cur_plan)[-1]
self.changes_of_plan.append([
f"{self.cur_item_id + 1} ({self._get_cur_item('name')})",
self._change_of_plan(old_plan_dict, new_plan_dict),
json.dumps(new_plan_dict)
])
self.plan_instruct = instruct_replan
self.cur_plan = result
self.withdraw = False
self.bid_history = [] # clear bid history
self.cur_item_id += 1
self.dialogue_history += [
replan_msg,
AIMessage(content=result),
]
self.llm_prompt_history.append({
'messages': [{x.type: x.content} for x in messages],
'result': result,
'tag': f'plan_{self.cur_item_id}'
})
if self.verbose:
print(get_colored_text(instruct_replan, 'blue'))
print(get_colored_text(result, 'green'))
print(f"Replan: {self.name} ({self.model_name}).")
return result
def _change_of_plan(self, old_plan: dict, new_plan: dict):
for k in new_plan:
if new_plan[k] != old_plan.get(k, None):
return True
return False
# *********** Belief Tracking and Sanity Check *********** #
def bid_sanity_check(self, bid_price, prev_round_max_bid, min_markup_pct):
# can't bid more than budget or less than previous highest bid
if bid_price < 0:
msg = None
else:
min_bid_increase = int(min_markup_pct * self._get_cur_item('price'))
if bid_price > self.budget:
msg = f"you don't have insufficient budget (${self.budget} left)"
elif bid_price < self._get_cur_item('price'):
msg = f"your bid is lower than the starting bid (${self._get_cur_item('price')})"
elif bid_price < prev_round_max_bid + min_bid_increase:
msg = f"you must advance previous highest bid (${prev_round_max_bid}) by at least ${min_bid_increase} ({int(100 * min_markup_pct)}%)."
else:
msg = None
return msg
def rebid_for_failure(self, fail_instruct: str):
result = self.bid(fail_instruct)
self.failed_bid_cnt += 1
return result
def _sanity_check_status_json(self, data: dict):
if data == {}:
return "Error: No parsible JSON in your response. Possibly due to missing a closing curly bracket '}', or unpasible values (e.g., 'profit': 1000 + 400, instead of 'profit': 1400)."
# Check if all expected top-level keys are present
expected_keys = ["remaining_budget", "total_profits", "winning_bids"]
for key in expected_keys:
if key not in data:
return f"Error: Missing '{key}' field in the status JSON."
# Check if "remaining_budget" is a number
if not isinstance(data["remaining_budget"], (int, float)):
return "Error: 'remaining_budget' should be a number, and only about your remaining budget."
# Check if "total_profits" is a dictionary with numbers as values
if not isinstance(data["total_profits"], dict):
return "Error: 'total_profits' should be a dictionary of every bidder."
for bidder, profit in data["total_profits"].items():
if not isinstance(profit, (int, float)):
return f"Error: Profit for {bidder} should be a number."
# Check if "winning_bids" is a dictionary and that each bidder's entry is a dictionary with numbers
if not isinstance(data["winning_bids"], dict):
return "Error: 'winning_bids' should be a dictionary."
for bidder, bids in data["winning_bids"].items():
if not isinstance(bids, dict):
return f"Error: Bids for {bidder} should be a dictionary."
for item, amount in bids.items():
if not isinstance(amount, (int, float)):
return f"Error: Amount for {item} under {bidder} should be a number."
# If everything is fine
return ""
def _status_json_to_text(self, data: dict):
if 'rule' in self.model_name: return ''
# Extract and format remaining budget
structured_text = f"* Remaining Budget: ${data.get('remaining_budget', 'unknown')}\n\n"
# Extract and format total profits for each bidder
structured_text += "* Total Profits:\n"
if data.get('total_profits'):
for bidder, profit in data['total_profits'].items():
structured_text += f" * {bidder}: ${profit}\n"
# Extract and list the winning bids for each item by each bidder
structured_text += "\n* Winning Bids:\n"
if data.get('winning_bids'):
for bidder, bids in data['winning_bids'].items():
structured_text += f" * {bidder}:\n"
if bids:
for item, amount in bids.items():
structured_text += f" * {item}: ${amount}\n"
else:
structured_text += f" * No winning bids\n"
return structured_text.strip()
def _belief_tracking(self, status_text: str):
'''
Parse status quo and check if the belief is correct.
'''
belief_json = extract_jsons_from_text(status_text)[-1]
# {"remaining_budget": 8000, "total_profits": {"Bidder 1": 1300, "Bidder 2": 1800, "Bidder 3": 0}, "winning_bids": {"Bidder 1": {"Item 2": 1200, "Item 3": 1000}, "Bidder 2": {"Item 1": 2000}, "Bidder 3": {}}}
budget_belief = belief_json['remaining_budget']
profits_belief = belief_json['total_profits']
winning_bids = belief_json['winning_bids']
msg = ''
# track belief of budget
self.total_self_belief_cnt += 1
if budget_belief != self.budget:
msg += f'- Your belief of budget is wrong: you have ${self.budget} left, but you think you have ${budget_belief} left.\n'
self.self_belief_error_cnt += 1
self.budget_error_history.append([
self._get_cur_item('name'),
budget_belief,
self.budget,
])
# track belief of profits
for bidder_name, profit in profits_belief.items():
if self.all_bidders_status.get(bidder_name) is None:
# due to a potentially unreasonable parsing
continue
if self.name in bidder_name:
bidder_name = self.name
self.total_self_belief_cnt += 1
else:
self.total_other_belief_cnt += 1
real_profit = self.all_bidders_status[bidder_name]['profit']
if profit != real_profit:
if self.name == bidder_name:
self.self_belief_error_cnt += 1
else:
self.other_belief_error_cnt += 1
msg += f'- Your belief of total profit of {bidder_name} is wrong: {bidder_name} has earned ${real_profit} so far, but you think {bidder_name} has earned ${profit}.\n'
# add to history
self.profit_error_history.append([
f"{bidder_name} ({self._get_cur_item('name')})",
profit,
real_profit
])
# track belief of winning bids
for bidder_name, items_won_dict in winning_bids.items():
if self.all_bidders_status.get(bidder_name) is None:
# due to a potentially unreasonable parsing
continue
real_items_won = self.all_bidders_status[bidder_name]['items_won']
# items_won = [(item, bid_price), ...)]
items_won_list = list(items_won_dict.keys())
real_items_won_list = [str(x) for x, _ in real_items_won]
if self.name in bidder_name:
self.total_self_belief_cnt += 1
else:
self.total_other_belief_cnt += 1
if not item_list_equal(items_won_list, real_items_won_list):
if bidder_name == self.name:
self.self_belief_error_cnt += 1
_bidder_name = f'you'
else:
self.other_belief_error_cnt += 1
_bidder_name = bidder_name
msg += f"- Your belief of winning items of {bidder_name} is wrong: {bidder_name} won {real_items_won}, but you think {bidder_name} won {items_won_dict}.\n"
self.win_bid_error_history.append([
f"{_bidder_name} ({self._get_cur_item('name')})",
', '.join(items_won_list),
', '.join(real_items_won_list)
])
return msg
def win_bid(self, item: Item, bid: int):
self.budget -= bid
self.profit += item.true_value - bid
self.items_won += [[item, bid]]
msg = f"Congratuations! You won {item} at ${bid}."# Now you have ${self.budget} left. Your total profit so far is ${self.profit}."
return msg
def lose_bid(self, item: Item):
return f"You lost {item}."# Now, you have ${self.budget} left. Your total profit so far is ${self.profit}."
# set the profit information of other bidders
def set_all_bidders_status(self, all_bidders_status: dict):
self.all_bidders_status = all_bidders_status.copy()
def set_withdraw(self, bid: int):
if bid < 0: # withdraw
self.withdraw = True
elif bid == 0: # enable discount and bid again
self.withdraw = False
else: # normal bid
self.withdraw = False
self.engagement_count += 1
self.engagement_history[self._get_cur_item('name')] += 1
# ****************** Logging ****************** #
# def _parse_hedging(self, plan: str): # deprecated
# prompt = PARSE_HEDGE_INSTRUCTION.format(
# item_name=self._get_cur_item(),
# plan=plan)
# with get_openai_callback() as cb:
# llm = ChatOpenAI(model='gpt-3.5-turbo-0613', temperature=0)
# result = llm([HumanMessage(content=prompt)]).content
# self.openai_cost += cb.total_cost
# # parse a number, which could be a digit
# hedge_percent = re.findall(r'\d+\.?\d*%', result)
# if len(hedge_percent) > 0:
# hedge_percent = hedge_percent[0].replace('%', '')
# else:
# hedge_percent = 0
# return float(hedge_percent)
def profit_report(self):
'''
Personal profit report at the end of an auction.
'''
msg = f"* {self.name}, starting with ${self.original_budget}, has won {len(self.items_won)} items in this auction, with a total profit of ${self.profit}.:\n"
profit = 0
for item, bid in self.items_won:
profit += item.true_value - bid
msg += f" * Won {item} at ${bid} over ${item.price}, with a true value of ${item.true_value}.\n"
return msg.strip()
def to_monitors(self, as_json=False):
# budget, profit, items_won, tokens
if len(self.items_won) == 0 and not as_json:
items_won = [['', 0, 0]]
else:
items_won = []
for item, bid in self.items_won:
items_won.append([str(item), bid, item.true_value])
profit_error_history = self.profit_error_history if self.profit_error_history != [] or as_json else [['', '', '']]
win_bid_error_history = self.win_bid_error_history if self.win_bid_error_history != [] or as_json else [['', '', '']]
budget_error_history = self.budget_error_history if self.budget_error_history != [] or as_json else [['', '']]
changes_of_plan = self.changes_of_plan if self.changes_of_plan != [] or as_json else [['', '', '']]
if as_json:
return {
'auction_hash': self.auction_hash,
'bidder_name': self.name,
'model_name': self.model_name,
'desire': self.desire,
'plan_strategy': self.plan_strategy,
'overestimate_percent': self.overestimate_percent,
'temperature': self.temperature,
'correct_belief': self.correct_belief,
'enable_learning': self.enable_learning,
'budget': self.original_budget,
'money_left': self.budget,
'profit': self.profit,
'items_won': items_won,
'tokens_used': self.llm_token_count,
'openai_cost': round(self.openai_cost, 2),
'failed_bid_cnt': self.failed_bid_cnt,
'self_belief_error_cnt': self.self_belief_error_cnt,
'other_belief_error_cnt': self.other_belief_error_cnt,
'failed_bid_rate': round(self.failed_bid_cnt / (self.total_bid_cnt+1e-8), 2),
'self_error_rate': round(self.self_belief_error_cnt / (self.total_self_belief_cnt+1e-8), 2),
'other_error_rate': round(self.other_belief_error_cnt / (self.total_other_belief_cnt+1e-8), 2),
'engagement_count': self.engagement_count,
'engagement_history': self.engagement_history,
'changes_of_plan': changes_of_plan,
'budget_error_history': budget_error_history,
'profit_error_history': profit_error_history,
'win_bid_error_history': win_bid_error_history,
'history': self.llm_prompt_history
}
else:
return [
self.budget,
self.profit,
items_won,
self.llm_token_count,
round(self.openai_cost, 2),
round(self.failed_bid_cnt / (self.total_bid_cnt+1e-8), 2),
round(self.self_belief_error_cnt / (self.total_self_belief_cnt+1e-8), 2),
round(self.other_belief_error_cnt / (self.total_other_belief_cnt+1e-8), 2),
self.engagement_count,
draw_plot(f"{self.name} ({self.model_name})", self.budget_history, self.profit_history),
changes_of_plan,
budget_error_history,
profit_error_history,
win_bid_error_history
]
def dialogue_to_chatbot(self):
# chatbot: [[Human, AI], [], ...]
# only dialogue will be sent to LLMs. chatbot is just for display.
assert len(self.dialogue_history) % 2 == 0
chatbot = []
for i in range(0, len(self.dialogue_history), 2):
# if exceeds the length of dialogue, append the last message
human_msg = self.dialogue_history[i].content
ai_msg = self.dialogue_history[i+1].content
if ai_msg == '': ai_msg = None
if human_msg == '': human_msg = None
chatbot.append([human_msg, ai_msg])
return chatbot
def draw_plot(title, hedge_list, profit_list):
x1 = [str(i) for i in range(len(hedge_list))]
x2 = [str(i) for i in range(len(profit_list))]
y1 = hedge_list
y2 = profit_list
fig, ax1 = plt.subplots()
color = 'tab:red'
ax1.set_xlabel('Bidding Round')
ax1.set_ylabel('Budget Left ($)', color=color)
ax1.plot(x1, y1, color=color, marker='o')
ax1.tick_params(axis='y', labelcolor=color)
for i, j in zip(x1, y1):
ax1.text(i, j, str(j), color=color)
ax2 = ax1.twinx()
color = 'tab:blue'
ax2.set_ylabel('Total Profit ($)', color=color)
ax2.plot(x2, y2, color=color, marker='^')
ax2.tick_params(axis='y', labelcolor=color)
for i, j in zip(x2, y2):
ax2.text(i, j, str(j), color=color)
lines1, labels1 = ax1.get_legend_handles_labels()
lines2, labels2 = ax2.get_legend_handles_labels()
ax2.legend(lines1 + lines2, labels1 + labels2, loc=0)
# fig.tight_layout()
plt.title(title)
return fig
def bidding_multithread(bidder_list: List[Bidder],
instruction_list,
func_type,
thread_num=5,
retry=1):
'''
auctioneer_msg: either a uniform message (str) or customed (list)
'''
assert func_type in ['plan', 'bid', 'summarize', 'replan']
result_queue = queue.Queue()
threads = []
semaphore = threading.Semaphore(thread_num)
def run_once(i: int, bidder: Bidder, auctioneer_msg: str):
try:
semaphore.acquire()
if func_type == 'bid':
result = bidder.bid(auctioneer_msg)
elif func_type == 'summarize':
result = bidder.summarize(auctioneer_msg)
elif func_type == 'plan':
result = bidder.init_plan(auctioneer_msg)
elif func_type == 'replan':
result = bidder.replan(auctioneer_msg)
else:
raise NotImplementedError(f'func_type {func_type} not implemented')
result_queue.put((True, i, result))
# except Exception as e:
# result_queue.put((False, i, str(trace_back(e))))
finally:
semaphore.release()
if isinstance(instruction_list, str):
instruction_list = [instruction_list] * len(bidder_list)
for i, (bidder, msg) in enumerate(zip(bidder_list, instruction_list)):
thread = threading.Thread(target=run_once, args=(i, bidder, msg))
thread.start()
threads.append(thread)
for thread in threads:
thread.join(timeout=600)
results = [result_queue.get() for _ in range(len(bidder_list))]
errors = []
for success, id, result in results:
if not success:
errors.append((id, result))
if errors:
raise Exception(f"Error(s) in {func_type}:\n" + '\n'.join([f'{i}: {e}' for i, e in errors]))
valid_results = [x[1:] for x in results if x[0]]
valid_results.sort()
return [x for _, x in valid_results]
def bidders_to_chatbots(bidder_list: List[Bidder], profit_report=False):
if profit_report: # usually at the end of an auction
return [x.dialogue_to_chatbot() + [[x.profit_report(), None]] for x in bidder_list]
else:
return [x.dialogue_to_chatbot() for x in bidder_list]
def create_bidders(bidder_info_jsl, auction_hash):
bidder_info_jsl = LoadJsonL(bidder_info_jsl)
bidder_list = []
for info in bidder_info_jsl:
info['auction_hash'] = auction_hash
bidder_list.append(Bidder.create(**info))
return bidder_list |