def eval_genome(genome, config):
net = neat.nn.FeedForwardNetwork.create(genome, config)
fitnesses = []
for runs in range(runs_per_net):
game = Game(20, 20)
# Run the given simulation for up to num_steps time steps.
fitness = 0.0
while True:
inputs = game.get_normalized_state()
action = net.activate(inputs)
# Apply action to the simulated snake
valid = game.step(np.argmax(action))
# Stop if the network fails to keep the snake within the boundaries or hits itself.
# The per-run fitness is the number of pills eaten
if not valid:
break
fitness = game.fitness
fitnesses.append(fitness)
# The genome's fitness is its worst performance across all runs.
return min(fitnesses)
def build_agent():
game = Game(
map_size=(20, 20),
initial_snake_length=3,
create_observation_strategy=InverseDistanceObservationStrategy,
create_reward_strategy=SquareExpRewardStrategy)
return snake.agent.Agent(env=game, hidden_nodes=[18, 18])
def eval_genome(genome, config):
net = neat.nn.FeedForwardNetwork.create(genome, config)
fitnesses = []
for runs in range(runs_per_net):
#pygame.init()
#screen = pygame.display.set_mode((20 * 16,20 * 16))
#screen.fill(pygame.Color('black'))
#pygame.display.set_caption('Snake')
#pygame.display.flip()
sim = Game(20, 20)
# Run the given simulation for up to num_steps time steps.
fitness = 0.0
while True:
inputs = sim.get_normalized_state()
action = net.activate(inputs)
# Apply action to the simulated snake
valid = sim.step(np.argmax(action))
# Stop if the network fails to keep the snake within the boundaries or hits itself.
# The per-run fitness is the number of pills eaten
if not valid:
break
fitness = sim.score
fitnesses.append(fitness)
# The genome's fitness is its worst performance across all runs.
return min(fitnesses)
def getTrainingData(self):
print('Getting Training Data . . .')
data = []
number = int(self.train_games / 20)
for x in range(self.train_games):
game = Game(x=self.x, y=self.y)
c_data = []
self.game = game
snake = game.start()
current_state = self.getState(snake)
for _ in range(self.max_steps):
action = self.getAction()
length = snake.length
done, snake, closer = game.step(action)
if done: break
elif not closer: continue
else:
correct_output = [0, 0, 0]
correct_output[action + 1] = 1
num = 1
if snake.length > length: num = 3
for _ in range(num):
c_data.append([current_state, correct_output])
current_state = self.getState(snake)
if snake.length > 2:
for el in c_data:
data.append(el)
if x % number == 0: print(f'{int(x/self.train_games*100)}%')
return data
def build_agent():
game = Game(
map_size=(20, 20),
create_observation_strategy=InverseDistanceObservationStrategy,
create_reward_strategy=SurvivalRewardStrategy)
game = MaxTimestepsWrapper(game, max_timesteps=1000)
return snake.agent.Agent(env=game, hidden_nodes=[18, 18])
def test_egg():
g = Game((5, 5), (10, 10))
# create a snake that takes almost all the space
g.snake = [ (i, j) for i in range(5) for j in range(5) ]
g.snake.remove( (2, 2))
assert g.random_egg() == (2, 2)
def __init__(self):
self.window = pygame.display.set_mode((270, 270))
self.game = Game(270, 270, 9, self.window, 0, 0)
self.model = self.load_model()
self.cnt = 0
if self.model != None:
self.game.agent.update_model(self.model)
self.game.agent.update_tgt(self.model)
def test_move():
g = Game((5, 5), (10, 10))
# create a small snake
g.snake = [ (1, 1), (2, 1), (3, 1) ]
g.egg = (0, 0)
g.move_snake(1, 0, False)
assert g.snake == [ (2, 1), (3, 1), (4, 1) ]
g.move_snake(1, 0, False)
assert g.snake == [ (3, 1), (4, 1), (0, 1) ]
# no u-turn
g.move_snake(-1, 0, False)
assert g.snake == [ (3, 1), (4, 1), (0, 1) ]
def showGame(self, model):
game = Game(x=self.x, y=self.y, gui=True)
self.game = game
while True:
snake = game.start()
steps = self.max_steps
current_state = self.getState(snake)
while True:
m = model.predict(np.array([current_state]))
action = list(m[0]).index(max(list(m[0]))) - 1
length = snake.length
done, snake, c = game.step(action)
if done: break
elif snake.length > length: steps = self.max_steps
else: current_state = self.getState(snake)
time.sleep(.05)
steps -= 1
if steps == 0:
break
def test(self, model):
print('Testing . . .')
num = int(self.test_games / 20)
lengths = []
game = Game(x=self.x, y=self.y)
self.game = game
for x in range(self.test_games):
snake = game.start()
steps = self.max_steps
current_state = self.getState(snake)
while True:
m = model.predict(np.array([current_state]))
action = list(m[0]).index(max(list(m[0]))) - 1
length = snake.length
done, snake, _ = game.step(action)
if done: break
elif snake.length > length: steps = self.max_steps
else: current_state = self.getState(snake)
steps -= 1
if steps == 0:
break
lengths.append(snake.length)
if x % num == 0: print(f'{int((x/self.test_games)*100)}%')
print(f'Average: {sum(lengths)/len(lengths)}')
from snake import Game, Renderer, KeyboardInput
H = 10
W = 10
game = Game(H, W)
renderer = Renderer(game)
input = KeyboardInput(renderer.window)
while True:
renderer.render_frame()
action = input.get_input()
if action:
game.input(action)
game.update()
if game.has_ended():
renderer.close_window()
print('THE END')
break
'''
try:
change = game.changed_tiles
renderer.render_frame(change)
action = input.get_input()
if action:
game.input(action)
game.update()
if game.has_ended():
print('THE END')
return self.env.outdim # define action-value table # number of states is: # # current value: 1-21 # # number of actions: # # Stand=0, Hit=1 av_table = ActionValueTable(27, 4) av_table.initialize(2.) game = Game() # define Q-learning agent learner = Q(0.5, 0.2) learner._setExplorer(EpsilonGreedyExplorer(0.0)) agent = LearningAgent(av_table, learner) # define the environment env = SnakeEnv(4, 27) env.init_game(15) # define the task task = BlackjackTask(env) # finally, define experiment experiment = Experiment(task, agent)
def __init__(self):
self.window = pygame.display.set_mode((270, 270))
self.game = Game(270, 270, 9, self.window, 0, 0)
self.model = self.load_model()
def __init__(self):
self.game = Game()
self.square_size = 9 # the observation size
self.timestep = 0
@property
def optimizer(self):
return self.q_network.optimizer
@property
def mask(self):
return self.q_network.mask
def __getattr__(self, name):
if name in self.hyper_params:
return self.hyper_params[name]
else:
raise AttributeError()
def __del__(self):
self.sess.close()
if __name__ == '__main__':
from snake import Game
from window import Window
number = 6
block_size = 20
g = Game(number=number)
window = Window(number=number,
block_size=block_size,
expansion=1.5,
speed=0.2)
dqn = DQN(game=g)
dqn.train(window=window)
f_x = int(0 if not food_relative[0] else 1 * np.sign(food_relative[0])) + 1 # Select food relative x
f_y = int(0 if not food_relative[1] else 1 * np.sign(food_relative[1])) + 1 # Select food relative y
for i, field in enumerate(view_area.ravel()):
if not field: # Ignore 0=Path
continue
add = (FIELD_STATES ** i) * field
discrete_index += add
return direction, f_x, f_y, discrete_index
if __name__ == "__main__":
game = Game(food_ammount=1, render=True)
valid = True
observation = Game().reset()
score = 0
q_table = np.load(f"{FILE}.npy", allow_pickle=True)
os.makedirs(f"{FILE}", exist_ok=True)
step = 0
while valid:
game.draw()
surface = pygame.display.get_surface()
pygame.image.save(surface, f"{FILE}/image_{step}.png")
old_observation = observation
current_q_values = get_discrete_vals(q_table, old_observation)
from snake import Game
# This is a sample Python script.
# Press Shift+F10 to execute it or replace it with your code.
# Press Double Shift to search everywhere for classes, files, tool windows, actions, and settings.
# Press the green button in the gutter to run the script.
if __name__ == '__main__':
snake = Game()
snake.run()
# See PyCharm help at https://www.jetbrains.com/help/pycharm/
if event.key not in KEY_TO_ACTION_MAP:
continue
# Act on bound keys.
observation, reward, done, info = env.step(
KEY_TO_ACTION_MAP[event.key])
print(
f"Observation: {observation}\tReward: {reward}\tDone: {done}\tInfo: {info}"
)
if done:
env.reset()
# Limit frame rate.
update_clock.tick(30)
do_game_loop()
pygame.quit()
if __name__ == "__main__":
from snake import Game
from snake.observation_strategies.default_observation_strategy import DefaultObservationStrategy
from snake.reward_strategies.default_reward_strategy import DefaultRewardStrategy
play(
Game(map_size=[10, 10],
initial_snake_length=3,
create_observation_strategy=DefaultObservationStrategy,
create_reward_strategy=DefaultRewardStrategy))
def test_game(self):
snake = Snake([[5, 4], [4, 4], [3, 4]])
game = Game(10, snake, Direction.UP)
# on se limite ici un peu artificiellement à des cellules carrées
parser = ArgumentParser()
parser.add_argument("-W",
"--width",
default=15,
type=int,
help="horizontal size of board, in cells")
parser.add_argument("-H",
"--height",
default=15,
type=int,
help="vertical size of board, in cells")
parser.add_argument("-C",
"--cell",
default=20,
type=int,
help="cell size in pixels")
args = parser.parse_args()
# dans l'objet args on va retrouver les 3 réglages
# tels que définis sur la ligne de commande
# par exemple
# args.width vaut l'entier 15 si on ne précise pas -W
# args.width vaut l'entier 20 si on précise -W 20
#
# on peut toujours invoquer le programme avec --help
# python main.py --help
print(args.width, args.height, args.cell)
game = Game((args.width, args.height), (args.cell, args.cell))
game.run()
def __init__(self):
self.action_space = Discrete(3) # 0 = turn left, 1 = do nothing, 2 = turn right
self.state = [0, 0, 1, 0]
self.game = Game()
self.reward = 0
self.done = False
def init_game(self, snake_size):
self.game = Game()
self.game.init_game(snake_size)
self.running = True
from snake import Game
if __name__ == "__main__":
a = Game('snake', 1280, 640, 10)
a.run()