def is_direction_blocked(snake: Snake, enemy: Snake, direction):
point = np.array(snake.get_head_position()) + np.array(direction)
#print(point)
return point.tolist() in snake.positions[:-1] \
or point.tolist() in enemy.positions \
or point[0] == 0 or point[1] == 0 \
or point[0] == SIZE or point[1] == SIZE
def enemyMovement(food: Food, enemy: Snake):
enemy_epsilon = 0.5
direction = 0
if np.random.random() <= enemy_epsilon:
direction = random.randint(0, 3)
else:
if food.position[0] < enemy.get_head_position()[0]:
direction = 2
elif food.position[0] > enemy.get_head_position()[0]:
direction = 3
elif food.position[1] < enemy.get_head_position()[1]:
direction = 0
elif food.position[1] > enemy.get_head_position()[1]:
direction = 1
else:
direction = random.randint(0, 3)
enemy.action([direction], "Enemy")
if obs_type == "Grid":
obs = getObsGrid(snakes, food, OBS_GRID_SIZE, fullGrid=False)
if obs_type == "Small":
obs = getObsSmall(snakes, food)
if player.reward == -DEATH_PENALTY:
done = True
info = ""
return obs, player.reward, done, info
score_history = []
score = 0
n_steps_history = []
for i in tqdm(range(num_episodes)):
player = Snake(0)
enemy = Snake(1)
enemy.positions = [(random.randint(0, SIZE - 1), random.randint(0, SIZE - 1))]
food = Food([player])
done = False
score = 0
# returns a numpy array of the state we care about
observation = getObsGrid(snakes=[player], food=food, size=OBS_GRID_SIZE, fullGrid=False)
#observation = getObsSmall([player, enemy], food)
n_steps = 0
while not done and n_steps < 100:
n_steps += 1
# action needs to be either 0,1,2 or 3
action = agentAC.choose_action(observation)
observation_, reward, done, info = step(snakes=[player, enemy], food=food, action=action, obs_type=obs_type)
agentAC.learn(observation, reward, observation_, done) # For Actor-Critic
# if start_q_table is None:
# q_table = {}
# for x1 in range(-SIZE+1, SIZE):
# for y1 in range(-SIZE+1, SIZE):
# q_table[(x1, y1)] = [np.random.uniform(-5, 0) for i in range(4)]
if start_q_table is not None:
print("Loaded q-table")
with open(start_q_table, "rb") as f:
q_table = pickle.load(f)
episode_rewards = []
episode_scores = []
for episode in range(HM_EPISODES):
player = Snake(0)
enemy = Snake(1)
food = Food([player, enemy])
show = False
if episode % SHOW_EVERY == 0:
print(f"on #{episode}, epsilon is {epsilon}")
print(
f"{SHOW_EVERY} ep mean: {np.mean(episode_rewards[-SHOW_EVERY:])}"
)
print(
f"{SHOW_EVERY} ep mean score: {np.mean(episode_scores[-SHOW_EVERY:])}"
)
print(episode_scores)
else:
show = False
def display():
pygame.init()
clock = pygame.time.Clock()
screen = pygame.display.set_mode((SCREEN_WIDTH, SCREEN_HEIGHT), 0, 32)
surface = pygame.Surface(screen.get_size())
surface = surface.convert()
drawGrid(surface)
player = Snake(0)
player.lives = 10
enemy = Snake(1)
enemy.positions = [(random.randint(0,
SIZE - 1), random.randint(0, SIZE - 1))]
food = Food([player])
myfont = pygame.font.SysFont("bahnschrift", 20)
iterations = 0
while (player.lives > 0):
for event in pygame.event.get():
if event.type == pygame.KEYDOWN:
if event.key == pygame.K_1:
print("SPace")
clock.tick(0)
#print(get_food_distance(player, food))
#getObsSmall([player, enemy], food)
clock.tick(1)
drawGrid(surface)
enemy.handle_keys()
obs = (player - food)
# no random action in display phase
action_space = np.array(q_table[obs]).copy()
player.action(action_space, "QL")
enemy.move()
handleSnakeCollisions(player, enemy)
handleFoodEating([player, enemy], food)
getObsGrid([player, enemy], food, size=7, fullGrid=True)
player.draw(surface)
enemy.draw(surface)
food.draw(surface)
screen.blit(surface, (0, 0))
text1 = myfont.render("Score AI {0}".format(player.score), 1,
(250, 250, 250))
text2 = myfont.render("Score Player {0}".format(enemy.score), 1,
(250, 250, 250))
screen.blit(text1, (5, 10))
screen.blit(text2, (SCREEN_WIDTH - 175, 10))
pygame.display.update()
print("Snake Player Final Score:", player.score)
def render(agent, obs_type: str, board_size: int):
pygame.init()
clock = pygame.time.Clock()
screen = pygame.display.set_mode((SCREEN_WIDTH, SCREEN_HEIGHT), 0, 32)
surface = pygame.Surface(screen.get_size())
surface = surface.convert()
drawGrid(surface)
player = Snake(0)
player.lives = 5
enemy = Snake(1)
enemy.lives = 5
enemy.positions = [[5,5]]
food = Food([player])
myfont = pygame.font.SysFont("bahnschrift", 20)
iterations = 0
while (player.lives > 4 and enemy.lives > 4):
clock.tick(4)
drawGrid(surface)
enemyMovement(food=food,enemy=enemy)
# no random action in display phase
if obs_type == "Grid":
obs = getObsGrid([player], food, size=board_size, fullGrid=False)
if obs_type == 'Small':
obs = getObsSmall([player, enemy], food)
action = agent.choose_action(obs)
action_space = [action]
player.action(action_space, "PG")
handleSnakeCollisions(player,enemy)
handleFoodEating([player, enemy], food)
player.draw(surface)
enemy.draw(surface)
food.draw(surface)
screen.blit(surface, (0, 0))
text1 = myfont.render("Score AI {0}".format(player.score), 1, (250, 250, 250))
#text2 = myfont.render("Score Player {0}".format(enemy.score), 1, (250, 250, 250))
screen.blit(text1, (5, 10))
#screen.blit(text2, (SCREEN_WIDTH - 175, 10))
pygame.display.update()
print("Snake Player Final Score:", player.score)
def get_enemy_direction_vector(snake: Snake, enemy: Snake):
return np.array(enemy.get_head_position()) - np.array(snake.get_head_position())
def get_food_direction_vector(snake: Snake, food: Food):
return np.array(food.position) - np.array(snake.get_head_position())
info = ""
return obs, player.reward, done, info
if __name__ == '__main__':
if agent_type == "AC":
agent = ActorCriticAgent(alpha=0.001, input_dims=6, gamma=0.99,
layer1_size=128, layer2_size=64, n_actions=4)
agent.actor_critic = T.load("model/AC-Small-5000")
score_history = []
score = 0
n_steps_history = []
for i in tqdm(range(num_episodes)):
player = Snake(0)
enemy = Snake(1)
food = Food([player, enemy])
done = False
score = 0
# returns a numpy array of the state we care about
if obs_type == "Grid":
observation = getObsGrid(snakes=[player], food=food, size=OBS_GRID_SIZE, fullGrid=False)
if obs_type == 'Small':
observation = getObsSmall([player, enemy], food)
n_steps = 0
while player.lives > 0 and enemy.lives > 0 and n_steps < 1000:
n_steps += 1
# action needs to be either 0,1,2 or 3
action = agent.choose_action(observation)
observation_, reward, done, info = step(snakes=[player, enemy], food=food, action=action)