np.random.seed(args.seed)
# create environment
env = gym.make(args.env)
env.seed(args.seed)
env.action_space.seed(args.seed)
train_tools.EVAL_SEED = args.seed
obs_dim = env.observation_space.shape[0]
act_dim = env.action_space.shape[0]
act_bound = env.action_space.high[0]
# create nets
actor_net = DDPGMLPActor(obs_dim=obs_dim,
act_dim=act_dim,
act_bound=act_bound,
hidden_size=[256, 256],
hidden_activation=nn.ReLU)
critic_net1 = MLPQsaNet(obs_dim=obs_dim,
act_dim=act_dim,
hidden_size=[256, 256],
hidden_activation=nn.ReLU)
critic_net2 = MLPQsaNet(obs_dim=obs_dim,
act_dim=act_dim,
hidden_size=[256, 256],
hidden_activation=nn.ReLU)
# create buffer
if args.show:
data_buffer = None
from common.networks import MLPQsaNet, DDPGMLPActor
if __name__ == '__main__':
# create environment
env = gym.make("Pendulum-v0")
# env = gym.make('LunarLanderContinuous-v2')
# env = gym.make('BipedalWalker-v3')
obs_dim = env.observation_space.shape[0]
act_dim = env.action_space.shape[0]
act_bound = env.action_space.high[0]
# create nets
actor_net = DDPGMLPActor(obs_dim=obs_dim,
act_dim=act_dim,
act_bound=act_bound,
hidden_size=[400, 300],
hidden_activation=nn.ReLU)
critic_net = MLPQsaNet(obs_dim=obs_dim,
act_dim=act_dim,
hidden_size=[400, 300],
hidden_activation=nn.ReLU)
# create optimizer
actor_optimizer = torch.optim.Adam(actor_net.parameters(), lr=1e-4)
critic_optimizer = torch.optim.Adam(critic_net.parameters(), lr=1e-3)
# create buffer
replay_buffer = ReplayBuffer(obs_dim=obs_dim,
act_dim=act_dim,