def makeMove(self, state):
# if there is just 1 action (PASS), avoid the comptutation
actions = self.getAllActions(state)
res_actions = []
if len(actions) == 1:
return actions[0]
# do the sampling x times, then pick most common action
x = 10
for i in xrange(x):
time_start = time.time()
cardsLeft = cards.diff(cards.allCards(),
[state.playedCards, self.hand])
otherRemaining = list(state.numRemaining)
del otherRemaining[self.idx]
hands = cards.dealHands(cardsLeft, otherRemaining)
hands.insert(self.idx, dict(self.hand))
root = mctsNode(state.playedCards, self.idx, hands, self.idx, None,
0, state.topCard, state.lastPlayed, state.finished)
loop_count = 0
while time.time() < time_start + budget:
loop_count += 1
nextNode = self.selection(root)
result = self.simulation(nextNode)
self.backpropagation(nextNode, result)
sorted_children = sorted(
root.children, key=lambda child: child.score / child.visits)
res_actions.append(sorted_children[-1].lastMove)
numActions = Counter(res_actions).most_common()
return numActions[0][0]
def makeMove(self, state):
# if there is just 1 action (PASS), avoid the comptutation
actions = self.getAllActions(state)
res_actions = []
if len(actions) == 1:
return actions[0]
# do the sampling x times, then pick most common action
x = 10
for i in xrange(x):
time_start = time.time()
cardsLeft = cards.diff(cards.allCards(), [state.playedCards, self.hand])
otherRemaining = list(state.numRemaining)
del otherRemaining[self.idx]
hands = cards.dealHands(cardsLeft , otherRemaining)
hands.insert(self.idx, dict(self.hand))
root = mctsNode(state.playedCards, self.idx, hands, self.idx,
None, 0, state.topCard, state.lastPlayed, state.finished)
loop_count = 0
while time.time() < time_start + budget:
loop_count += 1
nextNode = self.selection(root)
result = self.simulation(nextNode)
self.backpropagation(nextNode, result)
sorted_children = sorted(root.children, key = lambda child: child.score/child.visits)
res_actions.append(sorted_children[-1].lastMove)
numActions = Counter(res_actions).most_common()
return numActions[0][0]
def newGame(self, ordering):
deck = cards.allCards()
hands = cards.dealHands(deck, 52/self.numPlayers)
for i in xrange(self.numPlayers):
self.agents[i].hand = hands[i]
# swap cards
if ordering:
for i in xrange(2):
high = ordering.index(i)
low = ordering.index(self.numPlayers - i - 1)
cards.swapCards(hands[high], hands[low], 2 - i)
threes = collections.Counter()
for p, hand in enumerate(hands):
threes[p] += hand[0]
whosTurn = np.random.choice(cards.cardDictToList(threes))
playedCards = [cards.noCards() for i in xrange(self.numPlayers)]
self.initialState = state.State(playedCards, whosTurn, None, None, [])
def __init__(self, agents, hands=None, playedCards=None, whosTurn=None,
topCard=None, lastPlayed=None, finished=[]):
"""Initializes the game with the agents listed as the players.
:agents: Either a list of agent objects or a list of agent
constructors. If constructors, the dealing and ID assignment will be
done here; if objects, we simply assign the list to self.agents.
:hands: list of hand dicts; if supplied, will be used as the hands for
the agents.
:playedCards: list of played card dicts; if supplied, will be used for
the playedCards for the initialState.
:whosTurn: index of agent whose turn it is; if supplied, the game will
start with this player.
topCard, lastPlayed, and finished are all parameters passed on to the
initial state; for more information, see state.State's __init__.
"""
self.numPlayers = len(agents)
if hands is None:
deck = cards.allCards()
hands = cards.dealHands(deck, 52/self.numPlayers)
# if agents is a list of agent objects, set that to self.agents
if all(isinstance(a, agent.Agent) for a in agents):
self.agents = agents
# otherwise, construct the agents from the list of agent constructors
else:
self.agents = [agentConstructor(i, hand)
for i, (agentConstructor,hand) in
enumerate(zip(agents, hands))]
if whosTurn is None:
# randomly choose a player with a 3 to start
# (proportional to how many 3's they have)
threes = collections.Counter()
for p, hand in enumerate(hands):
threes[p] += hand[0]
whosTurn = np.random.choice(cards.cardDictToList(threes))
if playedCards is None:
playedCards = [cards.noCards() for i in xrange(self.numPlayers)]
self.initialState = state.State(playedCards, whosTurn, topCard,
lastPlayed, finished)
def simulate(args):
"""Function to simulate the other players' cards randomly and play out the
paranoid game tree based on those hands. Returns the best action.
"""
global nodesExpanded
nodesExpanded = 0
trial, state, index, hand = args
# subtract played cards and your own hand from cards remaining
cardsLeft = cards.diff(cards.allCards(), [state.playedCards, hand])
# get number of remaining cards for everyone else and deal hands
withoutMe = list(state.numRemaining)
del withoutMe[index]
hands = cards.dealHands(cardsLeft, withoutMe)
# put my hand back in
hands.insert(index, hand)
agents = map(lambda (i,h): ParanoidAgent(i, h),
zip(xrange(state.numPlayers), hands))
res = paranoid(state, 1, agents, -(sys.maxint -1), sys.maxint)
return res[0], nodesExpanded
def simulate(args):
"""Function to simulate the other players' cards randomly and play out the
paranoid game tree based on those hands. Returns the best action.
"""
global nodesExpanded
nodesExpanded = 0
trial, state, index, hand = args
# subtract played cards and your own hand from cards remaining
cardsLeft = cards.diff(cards.allCards(), [state.playedCards, hand])
# get number of remaining cards for everyone else and deal hands
withoutMe = list(state.numRemaining)
del withoutMe[index]
hands = cards.dealHands(cardsLeft, withoutMe)
# put my hand back in
hands.insert(index, hand)
agents = map(lambda (i, h): ParanoidAgent(i, h),
zip(xrange(state.numPlayers), hands))
res = paranoid(state, 1, agents, -(sys.maxint - 1), sys.maxint)
return res[0], nodesExpanded
def simulate(args):
"""Function to simulate the other players' cards randomly and play out the
max^n tree based on those hands. Returns the best action.
:trialNum: Trial number (for debugging and unique identification).
:node: The current State object.
:idx: The index of the current player.
:hand: The current player's hand, which is known.
:returns: The action tuple corresponding to the best action to take.
"""
trialNum, node, idx, hand = args
# subtract played cards and your own hand from cards remaining
cardsLeft = cards.diff(cards.allCards(), [node.playedCards, hand])
# get number of remaining cards for everyone else and deal hands
withoutMe = list(node.numRemaining)
del withoutMe[idx]
hands = cards.dealHands(cardsLeft, withoutMe)
# put my hand back in
hands.insert(idx, hand)
agents = map(lambda (i,h): MaxNAgent(i, h),
zip(xrange(node.numPlayers), hands))
bestAct, bestVal = maxN(node, agents, 0, 2*node.numPlayers)
return bestAct
import cards, dummyAgent, maxN, paranoid, mcts, state
from collections import Counter
cs = cards.allCards()
hand = Counter({0: 1, 5: 1, 7: 1, 10: 1, 12: 1})
playedCards = [
{c: 1 for c in xrange(13)} for i in xrange(3)
] + [cards.noCards() for i in xrange(0)]
node = state.State(
whosTurn=0,
playedCards=playedCards,
topCard = (1, 4),
)
dummy = dummyAgent.DummyAgent(0, hand)
maxn = maxN.MaxNAgent(0, hand)
para = paranoid.ParanoidAgent(0, hand)
mct = mcts.mctsAgent(0, hand)
readablehand = Counter({cards.cardRepr[k]: v for k, v in hand.iteritems()})
print 'Given hand:', readablehand
dumMove = list(dummy.makeMove(node))
dumMove[1] = cards.cardRepr[dumMove[1]]
movestr = lambda m: 'Play {} {}'.format(m[0], m[1])
print 'Initial move chosen by dummy:', movestr(dumMove)
maxMove = list(maxn.makeMove(node))
maxMove[1] = cards.cardRepr[maxMove[1]]
print 'Initial move chosen by max^n:', movestr(maxMove)
paraMove = list(para.makeMove(node))
