class MainApp(App):
def build(self):
#sound = SoundLoader.load('muzlome_Metallica_-_Sad_But_True_47954412.mp3')
#sound.play()
self.create_start_screen()
self.create_game_screen()
self.main_layout = FloatLayout()
self.main_layout.add_widget(self.start_screen)
return self.main_layout
def create_start_screen(self):
self.start_screen = FloatLayout(size_hint=(1, 1),
pos_hint={
'center_x': 0.5,
'center_y': 0.5
})
background_image = Image(source=BACKGROUND_IMAGE,
size_hint=[1, 1],
pos_hint={
'center_x': 0.5,
'center_y': 0.5
},
allow_stretch=True,
keep_ratio=False)
background_belki = Image(source=BELKI,
size_hint=[1, 1],
pos_hint={
'center_x': 0.5,
'center_y': 0.5
},
allow_stretch=True,
keep_ratio=False)
buttons = BoxLayout(orientation='vertical',
spacing=10,
size_hint=(0.3, 0.4),
pos_hint={
'center_x': 0.7,
'center_y': 0.4
})
start_game_button = Button(text='Start game',
on_press=self.reset_game_mode,
background_normal='',
background_color=BUTTON_COLOR,
color=[0, 0, 0, 1])
settings_button = Button(
text='Settings',
#on_press=None,
background_normal='',
background_color=BUTTON_COLOR,
color=[0, 0, 0, 1])
buttons.add_widget(start_game_button)
buttons.add_widget(settings_button)
self.start_screen.add_widget(background_image)
self.start_screen.add_widget(buttons)
def create_game_screen(self):
self.board = BoardWidget(FIELD_WIDTH, FIELD_HEIGHT)
exit_button = Button(text='Quit',
on_press=self.leave,
size_hint=[0.25, 0.1],
pos_hint={
'center_x': 1.30,
'center_y': 0.66
},
background_color=BUTTON_COLOR,
background_normal='')
restart_button = Button(text='Restart',
on_press=self.reset_game_mode,
size_hint=[0.25, 0.1],
pos_hint={
'center_x': 1.30,
'center_y': 0.90
},
background_color=BUTTON_COLOR,
background_normal='')
back_to_menu_button = Button(
text='Main menu', #on_press=self.back_to_main_menu,
size_hint=[0.25, 0.1],
pos_hint={
'center_x': 1.30,
'center_y': 0.78
},
background_color=BUTTON_COLOR,
background_normal='')
make_move_button = Button(text='Start game',
on_press=self.start_game,
size_hint=[0.25, 0.1],
pos_hint={
'center_x': 1.30,
'center_y': 0.4
},
background_color=BUTTON_COLOR,
background_normal='')
self.game_screen = FloatLayout(size_hint=[0.5, 1],
pos_hint={
'center_x': 0.5,
'center_y': 0.5
})
self.game_screen.add_widget(self.board)
self.game_screen.add_widget(exit_button)
self.game_screen.add_widget(restart_button)
self.game_screen.add_widget(back_to_menu_button)
self.game_screen.add_widget(make_move_button)
def reset_game_mode(self, _):
self.game = Game([(FirstStupidAlgorithm(), [2, 2])])
self.main_layout.remove_widget(self.main_layout.children[0])
self.main_layout.add_widget(self.game_screen)
def start_game(self, _): # FIXME not class method
Clock.schedule_interval(self.make_move, 0.1)
def make_move(self, _):
self.game.move()
self.board.refresh(self.game.board)
return not self.game.is_game_over()
def set_start_mode(self, _):
pass
def leave(self, button):
exit()
def training_loop(training_model, opponent_model, verbose=False):
winner = None
# for tensor board logging
log_dir = ("logs/fit/" + training_model._name +
datetime.datetime.now().strftime("%Y%m%d-%H%M%S"))
tensorboard = ModifiedTensorBoard(log_dir=log_dir)
# now execute the q learning
y = 0.9
eps = 0.5
interval_size = 500
num_episodes = interval_size * 1
decay_factor = (1000 * eps)**(
-1 / num_episodes
) # ensures that eps = 0.001 after `num_episodes` episodes
r_avg_list = []
sse_avg_list = []
wins = []
n_moves_list = []
moves_played = [0] * BOARD_SIZE
for i in tqdm(range(num_episodes), desc="Training"):
as_player = random.choice([0, 1])
eps *= decay_factor
g = Game(verbose=verbose)
if as_player == 1: # Training as player 1 so opponent makes first move
winner, board = g.move(opponent_model.move(g._board, 0))
else:
board = g._board
done = False
r_sum = 0
sse_sum = 0
move_num = 0
while not done:
random_move = False
move_num += 1
preds = training_model.predict(board, as_player)
# To encourage early exploration
if np.random.random() < eps:
move = np.random.randint(0, BOARD_SIZE - 1)
random_move = True
else:
move = training_model.move(board, as_player)
moves_played.append(move)
winner, new_board = g.move(move)
if winner is None:
opponent_move = opponent_model.move(new_board, 1 - as_player)
winner, new_board = g.move(opponent_move)
# Calculate reward amount
if winner == as_player:
done = True
wins.append(1)
r = 1000 - move_num**2
elif winner == 1 - as_player:
done = True
wins.append(0)
r = -(1000 - move_num**2)
elif winner == -1:
done = True
wins.append(None)
r = 1000
else:
r = move_num
if winner is None:
target = r + y * np.max(
training_model.predict(new_board, as_player))
else:
target = r
target_vec = deepcopy(preds[0])
target_vec[move] = target
training_model.fit_one(
as_player,
board,
np.array([target_vec]),
epochs=1,
verbose=0,
callbacks=[tensorboard],
)
new_preds = training_model.predict(board, as_player)
sse = sum([(x - y)**2 for x, y in zip(preds[0], target_vec)])
new_sse = sum([(x - y)**2
for x, y in zip(new_preds[0], target_vec)])
sse_sum += sse
if verbose:
print(f"""
{training_model._name} training as player: {as_player}, move: {move_num}, eps: {round(eps, 2)},
old preds: {[round(p, 2) for p in preds[0]]}, rand move: {random_move},
tgt preds: {[round(p, 2) for p in target_vec]}, reward: {r},
new preds: {[round(p, 2) for p in new_preds[0]]}, average last 20 games: {round(sum(r_avg_list[-20:])/20, 2)}
sse: {round(sse, 4)} >> {round(new_sse, 4)}
""")
board = new_board
r_sum += r
if verbose and ((i % interval_size == 0 and i > 0) or
(i == num_episodes - 1)):
run_game(training_model, opponent_model, verbose=True)
# Collect game level metrics
r_avg_list.append(round(r_sum, 2))
n_moves_list.append(move_num)
tensorboard.update_stats(reward_sum=r_sum,
wins=wins[-1],
n_moves_avg=n_moves_list[-1])
tensorboard.update_dist(moves_played=moves_played)
def main():
game = Game([(FirstStupidAlgorithm(), [6, 8])])
print(game.board)
while game.is_game_over() == False:
game.move()
print(game.board)
