args = get_ddpg_args_test()
experiment = args.experiment
seed = args.seed
policy_name = "DDPG"
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
file_name = "{}_{}_{}".format(policy_name, experiment, seed)
# Launch the env with our helper function
env = launch_env()
# Wrappers
env = ResizeWrapper(env)
env = NormalizeWrapper(env)
env = ImgWrapper(env) # to make the images from 160x120x3 into 3x160x120
env = ActionWrapper(env)
# env = DtRewardWrapper(env) # not during testing
state_dim = env.observation_space.shape
action_dim = env.action_space.shape[0]
max_action = float(env.action_space.high[0])
# Initialize policy
policy = DDPG(state_dim, action_dim, max_action, net_type="cnn")
policy.load(file_name, directory="./pytorch_models")
with torch.no_grad():
file_name = "{}_{}_{}".format(
policy_name,
experiment,
str(args.seed),
)
if not os.path.exists("./results"):
os.makedirs("./results")
if args.save_models and not os.path.exists("./pytorch_models"):
os.makedirs("./pytorch_models")
# Launch the env with our helper function
env = launch_env()
# Wrappers
env = ResizeWrapper(env)
env = NormalizeWrapper(env)
env = ImgWrapper(env) # to make the images from 160x120x3 into 3x160x120
env = ActionWrapper(env)
env = DtRewardWrapper(env)
# Set seeds
seed(args.seed)
state_dim = env.observation_space.shape
action_dim = env.action_space.shape[0]
max_action = float(env.action_space.high[0])
# Initialize policy
policy = DDPG(state_dim, action_dim, max_action, net_type="cnn")