def winner(self): """ Who won the game (if it's finished). :return: The (integer) id of the player who won if the game is finished (1 or 2). (None, None) if the game is not finished. """ winner = None points = None if self.__revoked is not None: # Thanks: Joshua Kenyon return util.other(self.__revoked), 3 if self.__p1_points >= 66: winner = 1 elif self.__p2_points >= 66: winner = 2 other_player_points = self.get_points(util.other(winner)) if other_player_points == 0: points = 3 elif other_player_points < 33: points = 2 else: points = 1 return winner, points
def heuristic_1a(self, player: int, depth: int,
curr_state: State) -> float:
if (curr_state.get_points(util.other(player)) +
curr_state.get_points(player)) == 0:
return 0
return curr_state.get_points(
util.other(player)) / (curr_state.get_points(util.other(player)) +
curr_state.get_points(player))
def __add_partial_trick_to_perspective(self, trick, player):
"""
Adds the card in the trick to the specified player's perspective
:param trick: A tuple signifying the trick which cards are revealed to the player
:param player: An integer signifying the player id
"""
if player == 1:
self.__p1_perspective[trick[util.other(player) - 1]] = "P2H"
else:
self.__p2_perspective[trick[util.other(player) - 1]] = "P1H"
def get_move(self, state):
# Find out which player we are
my_id = state.whose_turn()
# Our move: these will contain Planet objects
source = None
dest = None
source_strength = -1 # source score must end up as large as possible (start with very low value)
dest_strength = float(
'inf'
) # destination score must end up as small as possible (start with high value)
# Find my strongest planet (largest number of stationed ships)
for mine in state.planets(my_id):
strength = state.garrison(mine)
if strength > 1 and strength > source_strength:
source_strength = strength
source = mine
# Find the weakest enemy or neutral planet (smallest number of ships).
for his in (state.planets(u.other(my_id)) + state.planets(0)):
strength = state.garrison(his)
if strength < dest_strength:
dest_strength = strength
dest = his
if source is None or dest is None:
return None
return source.id(), dest.id()
def __evaluate_trick(self, trick):
"""
Evaluates who the winner of the specified trick is and returns it
:param trick: A tuple signifying the trick which is evaluated
:return: The winner's id as an integer
"""
if len(trick) != 2:
raise RuntimeError("Incorrect trick format. List of length 2 needed.")
if trick[0] is None or trick[1] is None:
raise RuntimeError("An incomplete trick was attempted to be evaluated.")
# If the two cards of the trick have the same suit
if Deck.get_suit(trick[0]) == Deck.get_suit(trick[1]):
# We only compare indices since the convention we defined in Deck
# puts higher rank cards at lower indices, when considering the same color.
return 1 if trick[0] < trick[1] else 2
if Deck.get_suit(trick[0]) == self.__deck.get_trump_suit():
return 1
if Deck.get_suit(trick[1]) == self.__deck.get_trump_suit():
return 2
# If the control flow has reached this point, the trick consists of two
# different non-trump cards. Since the new leader is determined by the
# output of this function, at this point the state object still considers
# it to be the non-leading player's turn. Thus, we determine that the winner
# is the other player, i.e. the leading player. Thanks: Daan Raven
return util.other(self.whose_turn())
def winning_game(self, state):
player = state. whose_turn()
opponent = util.other(player)
player_points = state.get_points(player)
opponent_points = state.get_points(opponent)
return opponent_points < player_points
def expand(self):
player = self.state.whose_turn()
moves = self.state.moves()
for move in moves:
new_state = self.state.next(move)
child_node = Node(new_state, self, move, [], util.other(player))
self.children.append(child_node)
def action_cost(self, player, depth, curr_state) -> float:
def backward_cost():
eval_vec = [self.heuristic_1a(player, depth, curr_state)]
return self.heuristics_eval(eval_vec)
def forward_cost():
eval_vec = [self.heuristic_2a(player, depth, curr_state)]
return self.heuristics_eval(eval_vec)
if self.__WIN_SCORE <= curr_state.get_points(util.other(player)):
return -1
elif self.__WIN_SCORE <= curr_state.get_points(player):
return 1
return (backward_cost() + forward_cost()) / 2
def action_cost(self, player, depth, state) -> float:
def backward_cost():
return (state.get_points(util.other(player))
) / self.__WIN_SCORE # the opponent's score + depth
def forward_cost():
return (
self.__WIN_SCORE - state.get_points(player)
) / self.__WIN_SCORE # my score --> underestimate, can be negative
if self.__WIN_SCORE <= state.get_points(util.other(player)):
return -1
elif self.__WIN_SCORE <= state.get_points(player):
return 1
return (backward_cost() + forward_cost()) / 2
def cant_win_trick(self, state):
player = state.whose_turn()
opponent = util.other(player)
leader = state.leader()
if opponent == leader:
opponents_played_card = state.get_opponents_played_card()
current_opponent_points = state.get_points(opponent)
current_points = state.get_points(player)
for move in state.moves():
next_state = state.next(move)
future_opponent_points = next_state.get_points(opponent)
future_points = next_state.get_points(player)
points_gained = future_points - future_opponent_points
if(points_gained <= 0): return True
return False
def features(state):
# type: (State) -> tuple[float, ...]
"""
Extract features from this state. Remember that every feature vector returned should have the same length.
:param state: A state to be converted to a feature vector
:return: A tuple of floats: a feature vector representing this state.
"""
feature_set = []
# Add player 1's points to feature set
p1_points = state.get_points(state.whose_turn())
# Add player 2's points to feature set
p2_points = state.get_points(util.other(state.whose_turn()))
# Add player 1's pending points to feature set
p1_pending_points = state.get_pending_points(state.whose_turn())
# Add plauer 2's pending points to feature set
p2_pending_points = state.get_pending_points(util.other(
state.whose_turn()))
# Get trump suit
trump_suit = state.get_trump_suit()
# Add phase to feature set
phase = state.get_phase()
# Add stock size to feature set
stock_size = state.get_stock_size()
# Add leader to feature set
leader = 1 if state.leader() == state.whose_turn() else 2
# Add whose turn it is to feature set
# whose_turn = state.whose_turn()
# Add opponent's played card to feature set
opponents_played_card = state.get_opponents_played_card()
################## You do not need to do anything below this line ########################
'''
correct the perspective, whether is 1 or 2
'''
player_hand = 'P1H'
player_win = 'P1W'
other_hand = 'P2H'
other_win = 'P2W'
if state.whose_turn() == 2:
player_hand = 'P2H'
player_win = 'P2W'
other_hand = 'P1H'
other_win = 'P1W'
perspective = state.get_perspective()
# Perform one-hot encoding on the perspective.
# Learn more about one-hot here: https://machinelearningmastery.com/how-to-one-hot-encode-sequence-data-in-python/
perspective = [
card if card != 'U' else [1, 0, 0, 0, 0, 0] for card in perspective
]
perspective = [
card if card != 'S' else [0, 1, 0, 0, 0, 0] for card in perspective
]
perspective = [
card if card != player_hand else [0, 0, 1, 0, 0, 0]
for card in perspective
]
perspective = [
card if card != other_hand else [0, 0, 0, 1, 0, 0]
for card in perspective
]
perspective = [
card if card != player_win else [0, 0, 0, 0, 1, 0]
for card in perspective
]
perspective = [
card if card != other_win else [0, 0, 0, 0, 0, 1]
for card in perspective
]
# Append one-hot encoded perspective to feature_set
feature_set += list(chain(*perspective))
# Append normalized points to feature_set
total_points = p1_points + p2_points
feature_set.append(p1_points / total_points if total_points > 0 else 0.)
feature_set.append(p2_points / total_points if total_points > 0 else 0.)
# Append normalized pending points to feature_set
total_pending_points = p1_pending_points + p2_pending_points
feature_set.append(
p1_pending_points /
total_pending_points if total_pending_points > 0 else 0.)
feature_set.append(
p2_pending_points /
total_pending_points if total_pending_points > 0 else 0.)
# Convert trump suit to id and add to feature set
# You don't need to add anything to this part
suits = ["C", "D", "H", "S"]
trump_suit_onehot = [0, 0, 0, 0]
trump_suit_onehot[suits.index(trump_suit)] = 1
feature_set += trump_suit_onehot
# Append one-hot encoded phase to feature set
feature_set += [1, 0] if phase == 1 else [0, 1]
# Append normalized stock size to feature set
feature_set.append(stock_size / 10)
# Append one-hot encoded leader to feature set
feature_set += [1, 0] if leader == 1 else [0, 1]
# Append one-hot encoded whose_turn to feature set
# feature_set += [1, 0] if whose_turn == 1 else [0, 1]
# Append one-hot encoded opponent's card to feature set
opponents_played_card_onehot = [
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
]
opponents_played_card_onehot[
opponents_played_card if opponents_played_card is not None else 20] = 1
feature_set += opponents_played_card_onehot
# Return feature set
return feature_set
def next(self,
move # type: tuple(int, int)
):
"""
Computes the next state based on the given move
:param move: Tuple of length 2 of which each element can either be an int or None
:return: Newly computed state based on current state and given move
"""
if self.__signature is not None and self.__signature != self.whose_turn():
raise RuntimeError('\n\nGame is in phase 1. Cannot view next state with imperfect information. Try making an assumption first.\n')
if self.finished():
raise RuntimeError('Gamestate is finished. No next states exist.')
# Start with a copy of the current state
state = self.clone() # type: State
# If we find an invalid move, we set the __revoked class variable
# To the pid of the player who made the incorrect move, and return the state as is.
if not state.__is_valid(move):
state.__revoked = state.whose_turn()
return state
# If move is a trump exchange
if move[0] is None:
# Store the indices we need in variables
trump_jack_index = move[1]
trump_card_index = state.__deck.get_trump_card_index()
# Perform trump jack exchange, perspective updated in function
state.__exchange_trump(trump_jack_index)
return state
# Change turns
state.__leads_turn = not state.__leads_turn
#Add the given move to the trick, store the whole trick in a variable
trick = state.__deck.set_trick(state.whose_turn(), move[0])
# At this point, we know that the move is not a trump jack exchange.
# Check if this move is a marriage
if move[1] is not None:
# A marriage cannot be melded by the non-leading player
if state.__leads_turn:
raise RuntimeError("Marriage was attempted to be melded by non-leading player")
# Update perspective since an additional card is revealed by the player who performs a marriage.
state.__deck.add_to_perspective(util.other(state.whose_turn()), move[1], "P" + str(state.whose_turn()) + "H")
# Trump suit marriage yields 40 points, regular yields 20, to be awarded at next trick win.
if Deck.get_suit(move[1]) == state.__deck.get_trump_suit():
state.__reserve_pending_points(state.whose_turn(), 40)
else:
state.__reserve_pending_points(state.whose_turn(), 20)
# If it is not the lead's turn, i.e. currently the trick is
# incomplete and we already know it's not a trump jack exchange
if not state.__leads_turn:
other = state.whose_turn()
state.__player1s_turn = not state.__player1s_turn
state.__deck.add_to_perspective(state.whose_turn(), trick[other-1], "P" + str(other) + "H")
return state
# At this point we know that it is the lead's turn and that a complete
# trick from the previous hand can be evaluated.
# Evaluate the trick and store the winner in the leader variable
leader = state.__evaluate_trick(trick)
state.__allocate_trick_points(leader, trick)
state.__deck.put_trick_away(leader)
if state.__phase == 2 and len(state.hand()) == 0 and not state.finished():
# If all cards are exhausted, the winner of the last trick wins the game
state.__set_points(leader, 66)
#Draw cards from stock
if state.__phase == 1:
state.__deck.draw_card(leader)
state.__deck.draw_card(util.other(leader))
if state.__deck.get_stock_size() == 0:
state.__phase = 2
# Set player1s_turn according to the leader variable
state.__player1s_turn = True if leader == 1 else False
# Returns state
return state
def get_opponents_played_card(self): """ :return: An integer representing the index of the card the opponent has played, None if no card played """ return self.__deck.get_trick()[util.other(self.whose_turn()) - 1]
def heuristic_3b(self, player: int, depth: int,
curr_state: State) -> float:
curr_state.get_points(player) / (curr_state.get_points(
util.other(player)) + curr_state.get_points(player))
def heuristic_1b(self, player: int, depth: int,
curr_state: State) -> float:
return curr_state.get_pending_points(util.other(player)) / (
curr_state.get_pending_points(util.other(player)) +
curr_state.get_pending_points(player))
dqn_number = random.choice([1, 2])
n_epi += 1
while not state.finished():
given_state = state.clone(
signature=state.whose_turn()) if state.get_phase() == 1 else state
action = (None, None)
n_steps += 1
if state.whose_turn() == dqn_number:
#print('ai')
action = e_greedy(given_state)
state = state.next(action)
while (state.whose_turn() == util.other(dqn_number)):
#print('o')
state2 = state.clone(signature=state.whose_turn()
) if state.get_phase() == 1 else state
if state.finished():
break
move = opponent.get_move(state2)
state = state.next(move)
reward = 0
if state.finished():
winner, score = state.winner()
reward = 1 if winner == dqn_number else 0
win_history.append(reward)
def evaluate(self, node):
p1_points = node.state.get_points(self.get_me())
p2_points = node.state.get_points(util.other(self.get_me()))
return 1 if p1_points - p2_points > 0 else 0
def value(self):
wins = self.outcome[self.parent.state.whose_turn()]
loses = self.outcome[util.other(self.parent.state.whose_turn())]
return wins - loses
def backward_cost():
return (state.get_points(util.other(player))
) / self.__WIN_SCORE # the opponent's score + depth