def __init__(self):
self.envs = gym.make('Walker2d-v2')
self.envs.seed(args.seed)
self.envpoch = 2048
self.inputsize = 17
self.actionsize = 6
self.net = PPOnet(self.inputsize, self.actionsize).double().to(device)
self.optimizer = optim.Adam(self.net.parameters(),
lr=args.lr,
eps=args.eps)
self.clip_param = 0.2
self.PPOepoch = args.ppoepoch
self.gamma = args.gamma
self.lam = args.gaelambda
self.out_record = None
self.trajectories = []
self.path_lsa = './csvfiles/lossa_lr' + str(
args.lr) + '_ppoepoch' + str(args.ppoepoch) + '_gamma' + str(
args.gamma) + '_gaelambda' + str(args.gaelambda) + '.csv'
self.path_lsv = './csvfiles/lossv_lr' + str(
args.lr) + '_ppoepoch' + str(args.ppoepoch) + '_gamma' + str(
args.gamma) + '_gaelambda' + str(args.gaelambda) + '.csv'
self.path_ep = './csvfiles/episode_lr' + str(
args.lr) + '_ppoepoch' + str(args.ppoepoch) + '_gamma' + str(
args.gamma) + '_gaelambda' + str(args.gaelambda) + '.csv'
self.scaler = utils.Scaler(17)
def __init__(self):
self.env = gym.make('Walker2d-v2')
#self.env.seed(args.seed)
self.envpoch = 2048
self.inputsize = 17
self.actionsize = 6
self.actor = Actor().double().to(device)
self.actor_target = Actor().double().to(device)
self.actor_optimizer = torch.optim.RMSprop(self.actor.parameters(), lr=1e-4)
self.critic = Critic().double().to(device)
self.critic_target = Critic().double().to(device)
self.critic_optimizer = torch.optim.RMSprop(self.critic.parameters(), lr=1e-3)
#self.random_noise = random_process.OrnsteinUhlenbeckActionNoise(self.actionsize)
self.clip_param = 0.2
self.PPOepoch = args.ppoepoch
self.gamma = args.gamma
self.lam = args.gaelambda
self.out_record = None
self.trainstep = 0
self.trajectories = []
self.path_lsa = './csvfiles/lossa_lr.csv'
self.path_lsv = './csvfiles/lossv_lr.csv'
self.path_ep = './csvfiles/episode_lr.csv'
self.scaler = utils.Scaler(17)
def __init__(self):
self.env = gym.make('Walker2d-v2')
self.envpoch = 1000
self.valuenet = valuenet().double().to(device)
self.policynet = policynet().double().to(device)
self.optimizer_value = optim.Adam(self.valuenet.parameters(),
lr=7e-5,
eps=1e-7)
self.optimizer_policy = optim.Adam(self.policynet.parameters(),
lr=7e-5,
eps=1e-7)
self.clip_param = 0.2
self.memory = store()
self.gamma = 0.99
self.lam = 0.95
self.out_record = None
self.trajectories = []
self.path_t7 = 'model_doublenet.t7'
self.path_lsa = "loss_doublenet_a.csv"
self.path_lsv = "loss_doublenet_v.csv"
self.path_ep = "episode_doublenet.csv"
self.scaler = utils.Scaler(18)
if os.path.isfile(self.path_t7):
self.net.load_state_dict(
torch.load(self.path_t7, map_location='cpu'))
def __init__(self):
self.env = gym.make('Walker2d-v2')
self.envpoch = 1000
self.net = PPOnet().double().to(device)
self.optimizer = optim.Adam(self.net.parameters(),
lr=args.lr,
eps=args.eps)
self.clip_param = 0.2
self.PPOepoch = args.ppoepoch
self.memory = store()
self.gamma = 0.99
self.lam = 0.95
self.out_record = None
self.trajectories = []
self.path_lsa = './csvfiles/lossa_lr' + str(
args.lr) + '_ppoepoch' + str(args.ppoepoch) + '.csv'
self.path_lsv = './csvfiles/lossv_lr' + str(
args.lr) + '_ppoepoch' + str(args.ppoepoch) + '.csv'
self.path_ep = './csvfiles/episode_lr' + str(
args.lr) + '_ppoepoch' + str(args.ppoepoch) + '.csv'
self.scaler = utils.Scaler(18)