Hanabi/Py-Hanabi
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

adjust greedy solver to hanabliveinstances. dump file

Browse source
This commit is contained in:
Maximilian Keßler 2023-05-19 12:18:38 +02:00
parent 6d19565f5e
commit 16cee69c82
Signed by: max
GPG key ID: BCC5A619923C0BA5
1 changed files with 28 additions and 18 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

44
greedy_solver.py
View file

@ -2,9 +2,11 @@
import collections import collections
import sys import sys
from enum import Enum from enum import Enum
from log_setup import logger
from time import sleep from time import sleep
from hanabi import DeckCard, Action, ActionType, GameState, HanabiInstance from hanabi import DeckCard, Action, ActionType, GameState, HanabiInstance
from hanab_live import HanabLiveInstance, HanabLiveGameState
from compress import link, decompress_deck from compress import link, decompress_deck
from database.database import conn from database.database import conn
@ -33,6 +35,7 @@ class CardState():
case CardType.Dispensable: case CardType.Dispensable:
return "Dispensable ({}) with weight {}".format(self.card, self.weight) return "Dispensable ({}) with weight {}".format(self.card, self.weight)
# TODO # TODO
def card_type(game_state, card): def card_type(game_state, card):
played = game_state.stacks[card.suitIndex] played = game_state.stacks[card.suitIndex]
@ -55,8 +58,8 @@ class GreedyStrategy():
self.earliest_draw_times.append([]) self.earliest_draw_times.append([])
for r in range(1, 6): for r in range(1, 6):
self.earliest_draw_times[s].append(max( self.earliest_draw_times[s].append(max(
game_state.deck.index(DeckCard(s,r)) - game_state.hand_size * game_state.num_players + 1, game_state.deck.index(DeckCard(s, r)) - game_state.hand_size * game_state.num_players + 1,
0 if r == 1 else self.earliest_draw_times[s][r - 2] 0 if r == 1 else self.earliest_draw_times[s][r - 2]
)) ))
# Currently, we do not add the time the 5 gets drawn to this, since this is rather a measurument on how # Currently, we do not add the time the 5 gets drawn to this, since this is rather a measurument on how
@ -64,7 +67,8 @@ class GreedyStrategy():
self.suit_badness = [sum(self.earliest_draw_times[s][:-1]) for s in range(0, game_state.num_suits)] self.suit_badness = [sum(self.earliest_draw_times[s][:-1]) for s in range(0, game_state.num_suits)]
def make_move(self): def make_move(self):
hand_states = [[CardState(card_type(self.game_state, card), card, None) for card in self.game_state.hands[p]] for p in range(self.game_state.num_players)] hand_states = [[CardState(card_type(self.game_state, card), card, None) for card in self.game_state.hands[p]]
for p in range(self.game_state.num_players)]
# find dupes in players hands, marke one card crit and the other one trash # find dupes in players hands, marke one card crit and the other one trash
p = False p = False
@ -95,7 +99,8 @@ class GreedyStrategy():
if state.card_type == CardType.Playable: if state.card_type == CardType.Playable:
copy_holders = set(self.game_state.holding_players(state.card)) copy_holders = set(self.game_state.holding_players(state.card))
copy_holders.remove(player) copy_holders.remove(player)
connecting_holders = set(self.game_state.holding_players(DeckCard(state.card.suitIndex, state.card.rank + 1))) connecting_holders = set(
self.game_state.holding_players(DeckCard(state.card.suitIndex, state.card.rank + 1)))
if len(copy_holders) == 0: if len(copy_holders) == 0:
# card is unique, imortancy is based lexicographically on whether somebody has the conn. card and the rank # card is unique, imortancy is based lexicographically on whether somebody has the conn. card and the rank
@ -123,15 +128,13 @@ class GreedyStrategy():
nextCopy = self.game_state.deck[self.game_state.progress:].index(card) nextCopy = self.game_state.deck[self.game_state.progress:].index(card)
except: except:
nextCopy = 1 nextCopy = 1
# state.weight = self.suit_badness[state.card.suitIndex] * nextCopy + 2 * (5 - state.card.rank) # state.weight = self.suit_badness[state.card.suitIndex] * nextCopy + 2 * (5 - state.card.rank)
state.weight = nextCopy + 2 * (5 - state.card.rank) state.weight = nextCopy + 2 * (5 - state.card.rank)
cur_hand = hand_states[self.game_state.turn] cur_hand = hand_states[self.game_state.turn]
plays = [cstate for cstate in cur_hand if cstate.card_type == CardType.Playable] plays = [cstate for cstate in cur_hand if cstate.card_type == CardType.Playable]
trash = next((cstate for cstate in cur_hand if cstate.card_type == CardType.Trash), None) trash = next((cstate for cstate in cur_hand if cstate.card_type == CardType.Trash), None)
# actual decision on what to do # actual decision on what to do
if len(plays) > 0: if len(plays) > 0:
@ -145,13 +148,14 @@ class GreedyStrategy():
dispensable = [cstate for cstate in cur_hand if cstate.card_type == CardType.Dispensable] dispensable = [cstate for cstate in cur_hand if cstate.card_type == CardType.Dispensable]
if len(dispensable) == 0: if len(dispensable) == 0:
self.game_state.in_lost_state = True self.game_state.in_lost_state = True
# raise ValueError("Lost critical card") # raise ValueError("Lost critical card")
else: else:
discard = min(dispensable, key=lambda s: s.weight) discard = min(dispensable, key=lambda s: s.weight)
self.game_state.discard(discard.card.deck_index) self.game_state.discard(discard.card.deck_index)
else: else:
self.game_state.clue() self.game_state.clue()
def test(): def test():
# seed p4v0s148 # seed p4v0s148
deck = decompress_deck("15wpspaodknlftabkpixbxiudqvrumhsgeakqucvgcrfmfhynwlj") deck = decompress_deck("15wpspaodknlftabkpixbxiudqvrumhsgeakqucvgcrfmfhynwlj")
@ -161,9 +165,9 @@ def test():
strat = GreedyStrategy(gs) strat = GreedyStrategy(gs)
while not gs.is_over(): while not gs.is_over():
strat.make_move() strat.make_move()
# print(strat.suit_badness) # print(strat.suit_badness)
# print(COLORS) # print(COLORS)
# strat.make_move() # strat.make_move()
print(gs.actions) print(gs.actions)
print(link(gs.to_json())) print(link(gs.to_json()))
@ -173,15 +177,18 @@ def test():
# crits = open("crits_lost.txt", "a") # crits = open("crits_lost.txt", "a")
def run_deck(seed, num_players, deck_str): def run_deck(seed, num_players, deck_str, variant_id):
deck = decompress_deck(deck_str) deck = decompress_deck(deck_str)
instance = HanabiInstance(deck, num_players) instance = HanabLiveInstance(deck, num_players, variant_id)
gs = GameState(instance) gs = HanabLiveGameState(instance)
strat = GreedyStrategy(gs) strat = GreedyStrategy(gs)
while not gs.is_over(): while not gs.is_over():
strat.make_move() strat.make_move()
if not gs.score == 25: if not gs.score == 25:
losses.write("{}-player seed {:10} {}:\n{}\n".format(num_players, seed, str(deck), link(gs))) logger.verbose(
"Greedy strategy lost {}-player seed {:10} {}:\n{}"
.format(num_players, seed, str(deck), link(gs))
)
return False return False
return True return True
@ -190,17 +197,20 @@ def run_samples(num_players, sample_size):
won = 0 won = 0
lost = 0 lost = 0
cur = conn.cursor() cur = conn.cursor()
cur.execute("SELECT seed, num_players, deck FROM seeds WHERE variant_id = 0 AND num_players = (%s) order by seed desc limit (%s)", (num_players, sample_size)) cur.execute(
"SELECT seed, num_players, deck, variant_id FROM seeds WHERE variant_id = 0 AND num_players = (%s) order by seed desc limit (%s)",
(num_players, sample_size))
for r in cur: for r in cur:
succ = run_deck(*r) succ = run_deck(*r)
if succ: if succ:
won += 1 won += 1
else: else:
lost += 1 lost += 1
print("won: {:4}, lost: {:4}".format(won, lost), end = "\r") print("won: {:4}, lost: {:4}".format(won, lost), end="\r")
print() print()
print("Total wins: {}%".format(round(100 * won / (lost + won), 2))) print("Total wins: {}%".format(round(100 * won / (lost + won), 2)))
if __name__ == "__main__": if __name__ == "__main__":
for p in range(2, 6): for p in range(2, 6):
print("Testing on {} players...".format(p)) print("Testing on {} players...".format(p))