from itertools import count
import time
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
checkpoint = torch.load('policy_net_with_tail.pth')
policy_net = convnet(config.BOARD_SIZE).float().to(device, non_blocking=True)
policy_net.load_state_dict(checkpoint['model_state_dict'])
policy_net.eval()
env = Snake(config.BOARD_SIZE)
while 1:
done = False
obs = env.reset()
cum_reward = 0
render = True
env.render()
for step in count(1):
action = select_action(obs, policy_net, 0, explore=False)
new_obs, reward, done = env.step(action)
cum_reward += reward
obs = new_obs
env.render()
if done:
cv2.destroyAllWindows()
break
print('Reward:', cum_reward)
def train(self, n, m, pars):
"""
(Game, int, int, dict()) -> None
train game and run each step as
sequence of frames
n: row tiles of the screen
m: col tiles of the screen
pars" parameters passed in for each processors
"""
# initialize
record = 0
game = Snake(n, m, pars.get('n_food', None))
agent = Agent(game, pars)
while True:
# get old state
state_old = game.get_state()
# get move
final_move = agent.get_action(state_old)
# perform move and get new state
reward, done, score = game.play_step(final_move, pars)
state_new = game.get_state()
# train short memory
agent.train_short_memory(state_old, final_move, reward, state_new,
done)
# remember
agent.remember(state_old, final_move, reward, state_new, done)
# end game if reached num_games from pars or DEFAULT_END_GAME_POINT
# if set to -1 then run for ever
if pars.get('num_games', DEFAULT_END_GAME_POINT) != -1:
if agent.n_games > pars.get('num_games',
DEFAULT_END_GAME_POINT):
quit()
break
# when game is over
if done:
# reset game attributes
# increase game generation
# train the long memory
game.reset()
agent.n_games += 1
agent.train_long_memory()
# new highscore
if score > record:
record = score
# save the best model_state
#agent.model.save()
# takes away food depending on given probability, up until 1 food remains
decrease_probability = pars.get('decrease_food_chance',
DECREASE_FOOD_CHANCE)
if (game.n_food > 1) and (random.random() <
decrease_probability):
game.n_food -= 1
# prints game information to console
print('Game', agent.n_games, 'Score', score, 'Record:', record)
# appends game information to txt filen at specified path
self.save_to_file(f"graphs/{pars.get('graph', 'test')}.txt",
agent.n_games, score, record)