def feedReward(self, game_result):
if game_result == 1:
reward = self.win_reward
elif game_result == 0:
reward = self.draw_reward
elif game_result == -1:
reward = self.lose_reward
if self.sign == State.X:
try:
#load states_value from a second QlearningPlayer
loaded_states_value = load(
"policy_3x3_100000_round_against_RandomPlayer_secondplayer"
)
except:
print("There's no proper file for see the future.")
else:
try:
#load states_value from a second QlearningPlayer
loaded_states_value = load(
"policy_3x3_100000_round_against_RandomPlayer_firstplayer")
except:
print("There's no proper file for see the future.")
states_rev = self.states[::-1]
for i in range(len(self.states)):
st = states_rev[i]
if self.states_value.get(st) is None:
self.states_value[st] = 0
if i != len(self.states) - 1:
st_before_hash = states_rev[i + 1]
st_before = self.getBoardStateFromHash(st_before_hash)
b = Board(self.board.size)
b.table = st_before
else:
b = Board(self.board.size)
possible_q_values = []
for poss_move in b.empty_cells():
b.table[poss_move[0]][poss_move[1]].state = self.rival_sign
possible_boardState_hash = self.getHash(b.table)
if self.sign == State.X:
try:
value = loaded_states_value.get(
possible_boardState_hash)
except:
value = None
else:
try:
value = loaded_states_value.get(
possible_boardState_hash)
except:
value = None
if value != None:
possible_q_values.append(value)
else:
possible_q_values.append(0)
possible_q_value = max(possible_q_values)
self.states_value[st] = (
1 - self.alpha) * self.states_value[st] + self.alpha * (
reward + self.decay_gamma * possible_q_value)
#reward = self.states_value[st]
self.states = []
def trainingClassifierPlayer(self,
X_file,
Y_Xmoves_file,
Y_Ymoves_file,
start_player,
boardSize,
classifier_num=1):
b = Board(boardSize)
X = load(X_file)
Y_Xmoves = load(Y_Xmoves_file)
Y_Ymoves = load(Y_Ymoves_file)
if start_player:
player1 = RandomForestClassifierPlayer(None, None, X, Y_Xmoves,
Y_Ymoves, classifier_num)
player2 = RandomPlayer()
testGame(player1, player2, b, 1000)
print("saving...")
player1.saveModels()
print("saved")
else:
player2 = RandomForestClassifierPlayer(None, None, X, Y_Xmoves,
Y_Ymoves, classifier_num)
player1 = RandomPlayer()
testGame(player1, player2, b, 1000)
print("saving...")
player2.saveModels()
print("saved")
def __init__(self, board):
BasePlayer.__init__(self, board)
X = load("X_random_random_secondplayer_4x4_10000game")
Y_Xmoves = load("YX_random_random_secondplayer_4x4_10000game")
Y_Ymoves = load("YY_random_random_secondplayer_4x4_10000game")
self.clf_x = self.training(X, np.ravel(Y_Xmoves))
self.clf_y = self.training(X, np.ravel(Y_Ymoves))
def __init__(self, clf_x_file, clf_y_file, X = [], Y_Xmoves = [], Y_Ymoves = [], classifier_num = 1):
BasePlayer.__init__(self)
self.classifier_num = classifier_num
if len(X) != 0 and len(Y_Xmoves) != 0 and len(Y_Ymoves) != 0:
self.clf_x = self.training(X, np.ravel(Y_Xmoves))
self.clf_y = self.training(X, np.ravel(Y_Ymoves))
elif clf_x_file != None and clf_y_file != None:
self.clf_x = load(clf_x_file)
self.clf_y = load(clf_y_file)
else:
raise Exception("You must train or load two ML model")
self.name = "RandomForestClassifierPlayer"
def __init__(self,
learning_file=None,
alpha=0.2,
epszilon=0.3,
gamma=0.9,
win_reward=1,
draw_reward=0.3,
lose_reward=0):
BasePlayer.__init__(self)
self.states = [] # record all positions taken
self.alpha = alpha
self.decay_gamma = gamma
self.win_reward = win_reward
self.draw_reward = draw_reward
self.lose_reward = lose_reward
self.name = "QLearningPlayer"
if learning_file == None:
self.states_value = {} # state -> value
self.exp_rate = epszilon
else:
self.states_value = load(learning_file)
self.exp_rate = 0.0