import pymc as mc import functools import itertools from collections import defaultdict as dd import pprint import progressbar def ResistanceGame(n_players): full_set = [("G1", True), ("G2", True), ("G3", True), ("E1", False), ("E2", False), ("G4", True), ("E3", False), ("G5", True), ("G6", True), ("E4", False)] return full_set[:n_players] class DeceptionGame(object): def __init__(self, player_array): self.player_array = player_array self.all_permutations = list(itertools.permutations(player_array)) self.n_players = len(player_array) self.trace = None self.observations = [] self.tid = 0 self.lady_will_duck = mc.Bernoulli("lady_will_duck", 0.5) self.mission_ducks_on_round = [None] * 5 self.mission_ducks_on_round[0] = mc.Bernoulli("missionduck_on_0", 0.5) self.mission_ducks_on_round[1] = mc.Bernoulli("missionduck_on_1", 0.5) self.mission_ducks_on_round[2] = mc.Bernoulli("missionduck_on_2", 0.5) self.mission_ducks_on_round[3] = mc.Bernoulli("missionduck_on_3", 0.0) self.mission_ducks_on_round[4] = mc.Bernoulli("missionduck_on_4", 0.0) self.ignorance_on_round = [None] * 5 self.ignorance_on_round[0] = mc.Bernoulli("ignorance_on_0", 0.9) self.ignorance_on_round[1] = mc.Bernoulli("ignorance_on_1", 0.7) self.ignorance_on_round[2] = mc.Bernoulli("ignorance_on_2", 0.5) self.ignorance_on_round[3] = mc.Bernoulli("ignorance_on_3", 0.3) self.ignorance_on_round[4] = mc.Bernoulli("ignorance_on_4", 0.3) def player_is_good(self, deal, player): return deal[player][1] def player_is_role(self, deal, player): return deal[player][0] def add_known_alliance(self, player_id, is_good): transaction = [] def obs(deal): if self.player_is_good(deal, player_id) == is_good: return True else: return None transaction.append(obs) self.observations.append(transaction) self.tid += 1 def add_known_role(self, player_id, role_str): transaction = [] def obs(deal): if self.player_is_role(deal, player_id) == role_str: return True else: return None transaction.append(obs) self.observations.append(transaction) self.tid += 1 def player_sees_player_and_claims(self, p1, p2, claim): transaction = [] def obs(deal): if self.player_is_good(deal, p1): if self.player_is_good(deal, p2) == claim: return True else: return None else: if self.lady_will_duck.rand(): if self.player_is_good(deal, p2) == claim: return True else: return False else: if self.player_is_good(deal, p2) == claim: return False else: return True transaction.append(obs) self.observations.append(transaction) self.tid += 1 def do_mission(self, team, fails, must_fail, r): transaction = [] rnd = r - 1 def obs(deal): n_actually_good_people = sum([int(self.player_is_good(deal, x)) for x in team]) n_spies = len(team) - n_actually_good_people if n_spies == 0: if fails != 0: return None else: return True else: if fails == 0: if must_fail: return None duck = False for i in range(n_spies - fails): duck = duck or self.mission_ducks_on_round[rnd].rand() return duck else: if fails > n_spies: return None return True transaction.append(obs) self.observations.append(transaction) self.tid += 1 def do_vote(self, team, votes, r): transaction = [] rnd = r - 1 def obs(deal): n_actually_good_people = sum([int(self.player_is_good(deal, x)) for x in team]) n_spies = len(team) - n_actually_good_people could_happen = True for player, vote in enumerate(votes): if self.player_is_good(deal, player): if player in team: continue elif n_spies > 0: if vote == 1: if self.ignorance_on_round[rnd].rand(): continue else: return False else: if vote == 0: if self.ignorance_on_round[rnd].rand(): continue else: return False else: if player in team: continue elif n_spies == 0: if vote == 1: if self.ignorance_on_round[rnd].rand(): continue else: return False else: if vote == 0: if self.ignorance_on_round[rnd].rand(): continue else: return False return could_happen transaction.append(obs) self.observations.append(transaction) self.tid += 1 def eval(self, length=10): deck = self.all_permutations[:] new_deck = [] trace = [] progress = progressbar.ProgressBar(widgets=["Simulating games: ", progressbar.Bar(marker="*"), " ", progressbar.ETA()]) for i in progress(range(length)): for deal in deck: f_list = [] for obs in self.observations: for tid in obs: f_list.append(tid) is_bad = False dont_copy = False for f in f_list: out = f(deal) if out is None: is_bad = True dont_copy = True break if out is True: continue if out is False: is_bad = True continue if not is_bad: trace.append(deal) if not dont_copy: new_deck.append(deal) deck = new_deck new_deck = [] self.trace = trace def report(self): if self.trace is None: self.eval() return self.get_player_data() def get_player_data(self): out = [] for i in range(self.n_players): out.append({}) out[i]["role"] = dd(float) out[i]["side"] = dd(float) progress = progressbar.ProgressBar(widgets=["Reticulating splines: ", progressbar.Bar(marker="*"), " ", progressbar.ETA()]) size = len(self.trace) * 1.0 for deal in progress(self.trace): for i, card in enumerate(deal): role, side = card out[i]["role"][role] += 1.0 / size out[i]["side"][side] += 1.0 / size for i in range(self.n_players): out[i]["role"] = dict(out[i]["role"]) out[i]["side"] = dict(out[i]["side"]) return out def _aggregate(self, l, i): out = dd(float) size = len(l) * 1.0 for deal in l: out[deal[i]] += 1 / size return dict(out) def print_report(self): pp = pprint.PrettyPrinter(indent = 4) pp.pprint(self.report()) base_game = DeceptionGame(ResistanceGame(5)) #base_game.add_known_role(0, "G1") #base_game.add_known_alliance(1, False) #base_game.player_sees_player_and_claims(0, 1, True) #base_game.player_sees_player_and_claims(1, 2, False) #base_game.player_sees_player_and_claims(2, 3, False) base_game.do_vote([1,2], [0,1,1,0,1], 1) base_game.do_mission([1,2], 0, False, 1) base_game.do_vote([0, 1, 2], [1,1,1,0,1], 2) base_game.do_mission([0, 1, 2], 1, False, 2) base_game.do_vote([3, 4], [0,0,1,1,1], 3) base_game.do_mission([3, 4], 0, False, 3) base_game.do_vote([3, 4], [0,0,1,1,1], 4) base_game.do_mission([0, 3, 4], 1, False, 4) #base_game.do_vote([1, 3, 4], [0,1,1,0,1], 5) #base_game.do_mission([1, 3, 4], 1, True, 5) base_game.eval(100) base_game.print_report()