def __init__(self):
self.env = flappy_bird_gym.make("FlappyBird-v0")
self.episodes = 1000
self.state_space = self.env.observation_space.shape[0]
self.action_space = self.env.action_space.n
self.memory = deque(maxlen=2000)
self.gamma = 0.95
self.epsilon = 1
self.epsilon_decay = 0.9999
self.epsilon_min = 0.01
self.batch_number = 64
self.train_start = 1000
self.jump_prob = 0.01
self.model = NeuralNetwork(input_shape=(self.state_space, ),
output_shape=self.action_space)
def __init__(self):
#env Variables
self.env = flappy_bird_gym.make("FlappyBird-v0")
self.episodes = 1000
self.state_space = self.env.observation_space.shape[0]
self.action_space = self.env.action_space.n
#HyperParamters
self.memory = deque(maxlen=2000)
self.gamma = 0.95 # this is the discount rate. Setting it to .95 prioritizes imediate rewards rather then future rewards.
self.epsilon = 1
self.epsilon_decay = 0.9999
self.epsilon_min = 0.01
self.batch_number = 64
self.train_start = 1000
self.jump_prob = 0.01
self.model = NeuralNetwork(input_shape=(self.state_space, ),
output_shape=self.action_space)
def __init__(self, params_file):
self.env = flappy_bird_gym.make("FlappyBird-v0")
self.state_space = self.env.observation_space.shape[0]
self.action_space = self.env.action_space.n
self.params_file = params_file
self.params = self.read_params()
self.memory = deque(maxlen=2000)
self.episodes = self.params['training']['num_episodes']
self.gamma = self.params['training']['discount_rate']
self.epsilon = self.params['training']['init_exploration_rate']
self.epsilon_decay = self.params['training']['exploration_decay_rate']
self.epsilon_min = self.params['training']['min_exploration_rate']
self.batch_number = 64 # 16, 32, 64, 128
self.train_start = 1000
self.jump_prob = 0.01
self.model = NeuralNet(input_shape=(self.state_space, ),
output_shape=self.action_space)
def __init__(self):
self.env = flappy_bird_gym.make('FlappyBird-v0')
self.episodes = 1000
self.state_space = self.env.observation_space.shape[0]
self.action_space = self.env.action_space.n
self.memory = deque(maxlen=2000)
#Hyperparameters
#gives more wight on imediate rewards
self.gamma = 0.95
#probability of taking a random action
self.epsilon = 1
self.epsilon_decay = 0.9999
#once we decay epsilon, probability of takind a random action
self.epsilon_min = 0.01
#amount of datapoints added to neural network
self.batch_number = 64
self.train_start = 1000
self.jump_prob = 0.01
self.model = NeuralNetwork(input_shape=(self.state_space, ),
output_shape=self.action_space)
action='store_true',
help='use the RGB version')
parser.add_argument('-sd',
dest='subdir',
metavar='SUBDIR',
default='',
type=str,
help='subdir for weights')
args = parser.parse_args()
return args
args = parse_args()
birds = BirdsPopulation(rgb=args.rgb)
env_str = f"FlappyBird{'-rgb-' if args.rgb else ''}-v0"
envs = {
bird.id: flappy_bird_gym.make(env_str)
for bird in birds.population
}
main_env = flappy_bird_gym.make(env_str)
if args.initial_gen > 0 or args.load:
birds.load_population(generation=args.initial_gen, subdir=args.subdir)
if args.train:
evolution(args.initial_gen)
elif not args.plot:
bird = birds.select_fittest(1)
print(
f'Running single run using BIRD-{bird.id} from generation #{birds.generation}'
)
run_with_render(bird, recording=True)
else:
# trainer.restore("/home/edoardo/ray_results/tune_PPO_lunar_hover/PPO_LunarHover_7ba4e_00000_0_2021-04-02_19-01-43/checkpoint_990/checkpoint-990")
trainer.restore(
"/home/edoardo/ray_results/tune_PPO_flappy_bird/PPO_my_env_95b43_00000_0_2021-04-03_14-36-24/checkpoint_2280/checkpoint-2280"
)
policy = trainer.get_policy()
# sequential_nn = convert_ray_simple_policy_to_sequential(policy).cpu()
sequential_nn = convert_ray_policy_to_sequential(policy).cpu()
# l0 = torch.nn.Linear(4, 2, bias=False)
# l0.weight = torch.nn.Parameter(torch.tensor([[0, 0, 1, 0], [0, 0, 0, 1]], dtype=torch.float32))
# layers = [l0]
# for l in sequential_nn:
# layers.append(l)
# nn = torch.nn.Sequential(*layers)
nn = sequential_nn
env = flappy_bird_gym.make("FlappyBird-v0")
# env.render()
plot_index = 1
position_list = []
# env.render()
n_trials = 10
cumulative_reward = 0
clock = pygame.time.Clock()
for i in range(n_trials):
state = env.reset()
# env.state[2] = 0.01
# env.state[2] = 0.045
# env.state[3] = -0.51
# state = np.array(env.state)
state_np = np.array(state)
print(state_np)
def env_creator(env_config):
import flappy_bird_gym
env = flappy_bird_gym.make("FlappyBird-v0")
return env