def __init__(self, state_size, action_size):
self.t = 0
self.max_Q = 0
self.train = True
self.lane_detection = False # Set to True to train on images with segmented lane lines
# Get size of state and action
self.state_size = state_size
self.action_size = action_size
# These are hyper parameters for the DQN
self.discount_factor = 0.99
self.learning_rate = 1e-4
if (self.train):
self.epsilon = 1.0
self.initial_epsilon = 1.0
else:
self.epsilon = 1e-6
self.initial_epsilon = 1e-6
self.epsilon_min = 0.02
self.batch_size = 64
self.train_start = 100
self.explore = 10000
# Create replay memory using deque
self.memory = deque(maxlen=100000)
self.vae_used = False
if self.vae_used:
vae_path = 'vae-level-0-dim-32.pkl'
# assert vae_path == None , "Missing weight file for vae"
self.vae = load_vae(vae_path)
# Create main model and target model
self.model = self.build_model()
self.target_model = self.build_model()
# Copy the model to target model
# --> initialize the target model so that the parameters of model & target model to be same
self.update_target_model()
'{}_{}'.format(ENV_ID, args.exp_id))
else:
log_path = os.path.join(folder, algo)
model_path = "{}/{}.pkl".format(log_path, ENV_ID)
assert os.path.isdir(log_path), "The {} folder was not found".format(
log_path)
assert os.path.isfile(
model_path), "No model found for {} on {}, path: {}".format(
algo, ENV_ID, model_path)
model = ALGOS[algo].load(model_path)
if args.vae_path != '':
print("Loading VAE ...")
vae = load_vae(args.vae_path)
if vae is None:
N_COMMAND_HISTORY = 0
env = DonkeyVAEEnv(level=LEVEL,
frame_skip=TEST_FRAME_SKIP,
vae=vae,
const_throttle=None,
min_throttle=MIN_THROTTLE,
max_throttle=MAX_THROTTLE,
max_cte_error=10,
n_command_history=N_COMMAND_HISTORY)
env = Recorder(env, folder=args.record_folder, verbose=1)
try:
env = TeleopEnv(env, model=model)
args = parser.parse_args()
set_global_seeds(args.seed)
if args.trained_agent != "":
assert args.trained_agent.endswith('.pkl') and os.path.isfile(args.trained_agent), \
"The trained_agent must be a valid path to a .pkl file"
tensorboard_log = None if args.tensorboard_log == '' else args.tensorboard_log + '/' + ENV_ID
print("=" * 10, ENV_ID, args.algo, "=" * 10)
vae = None
if args.vae_path != '':
print("Loading VAE ...")
vae = load_vae(args.vae_path)
elif args.random_features:
print("Randomly initialized VAE")
vae = load_vae(z_size=Z_SIZE)
# Save network
args.save_vae = True
else:
print("Learning from pixels...")
# Load hyperparameters from yaml file
with open('hyperparams/{}.yml'.format(args.algo), 'r') as f:
hyperparams = yaml.load(f, Loader=yaml.UnsafeLoader)[BASE_ENV]
hyperparams['seed'] = args.seed
# Sort hyperparams that will be saved
saved_hyperparams = OrderedDict([(key, hyperparams[key])
args = parser.parse_args()
set_global_seeds(args.seed)
if args.trained_agent != "":
assert args.trained_agent.endswith('.pkl') and os.path.isfile(args.trained_agent), \
"The trained_agent must be a valid path to a .pkl file"
tensorboard_log = None if args.tensorboard_log == '' else os.path.join(args.tensorboard_log, ID_ENV)
print("=" * 10, ID_ENV, args.algo, "=" * 10)
vae = None
if args.vae_path != '':
print("Loading VAE ...")
vae = load_vae(args.vae_path, input_dimension=(IMAGE_HEIGHT, IMAGE_WIDTH, 3))
elif args.random_features:
print("Randomly initialized VAE")
vae = load_vae(z_size=Z_SIZE)
# Save network
args.save_vae = True
else:
print("Learning from pixels...")
# Load hyperparameters from yaml file
with open('hyperparams/{}.yml'.format(args.algo), 'r') as f:
hyperparams = yaml.load(f)[args.name_env] # wyb
# Sort hyperparams that will be saved
saved_hyperparams = OrderedDict([(key, hyperparams[key]) for key in sorted(hyperparams.keys())])
# save vae path
