bacow/easy-rl
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update rainbowdqn

Browse Source
This commit is contained in:
johnjim0816
2022-05-31 01:20:58 +08:00
parent cfc0f6492e
commit c7c94468c9
149 changed files with 1866 additions and 1549 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
codes/TD3/agent.py
View File

@@ -21,8 +21,8 @@ class Actor(nn.Module):
'''[summary]
Args:
input_dim (int): 输入维度，这里等于state_dim
output_dim (int): 输出维度，这里等于action_dim
input_dim (int): 输入维度，这里等于n_states
output_dim (int): 输出维度，这里等于n_actions
max_action (int): action的最大值
'''
super(Actor, self).__init__()

8
codes/TD3/memory.py
View File

@@ -14,13 +14,13 @@ import torch
class ReplayBuffer(object):
def __init__(self, state_dim, action_dim, max_size=int(1e6)):
def __init__(self, n_states, n_actions, max_size=int(1e6)):
self.max_size = max_size
self.ptr = 0
self.size = 0
self.state = np.zeros((max_size, state_dim))
self.action = np.zeros((max_size, action_dim))
self.next_state = np.zeros((max_size, state_dim))
self.state = np.zeros((max_size, n_states))
self.action = np.zeros((max_size, n_actions))
self.next_state = np.zeros((max_size, n_states))
self.reward = np.zeros((max_size, 1))
self.not_done = np.zeros((max_size, 1))
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

6
codes/TD3/task0_eval.py
View File

@@ -74,10 +74,10 @@ if __name__ == "__main__":
env.seed(cfg.seed) # Set seeds
torch.manual_seed(cfg.seed)
np.random.seed(cfg.seed)
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.shape[0]
n_states = env.observation_space.shape[0]
n_actions = env.action_space.shape[0]
max_action = float(env.action_space.high[0])
td3= TD3(state_dim,action_dim,max_action,cfg)
td3= TD3(n_states,n_actions,max_action,cfg)
cfg.model_path = './TD3/results/HalfCheetah-v2/20210416-130341/models/'
td3.load(cfg.model_path)
td3_rewards,td3_ma_rewards = eval(cfg.env,td3,cfg.seed)

12
codes/TD3/task0_train.py
View File

@@ -72,7 +72,7 @@ def train(cfg,env,agent):
else:
action = (
agent.choose_action(np.array(state))
+ np.random.normal(0, max_action * cfg.expl_noise, size=action_dim)
+ np.random.normal(0, max_action * cfg.expl_noise, size=n_actions)
).clip(-max_action, max_action)
# Perform action
next_state, reward, done, _ = env.step(action)
@@ -121,11 +121,11 @@ def train(cfg,env,agent):
# else:
# action = (
# agent.choose_action(np.array(state))
# + np.random.normal(0, max_action * cfg.expl_noise, size=action_dim)
# + np.random.normal(0, max_action * cfg.expl_noise, size=n_actions)
# ).clip(-max_action, max_action)
# # action = (
# # agent.choose_action(np.array(state))
# # + np.random.normal(0, max_action * cfg.expl_noise, size=action_dim)
# # + np.random.normal(0, max_action * cfg.expl_noise, size=n_actions)
# # ).clip(-max_action, max_action)
# # Perform action
# next_state, reward, done, _ = env.step(action)
@@ -157,10 +157,10 @@ if __name__ == "__main__":
env.seed(cfg.seed) # Set seeds
torch.manual_seed(cfg.seed)
np.random.seed(cfg.seed)
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.shape[0]
n_states = env.observation_space.shape[0]
n_actions = env.action_space.shape[0]
max_action = float(env.action_space.high[0])
agent = TD3(state_dim,action_dim,max_action,cfg)
agent = TD3(n_states,n_actions,max_action,cfg)
rewards,ma_rewards = train(cfg,env,agent)
make_dir(cfg.result_path,cfg.model_path)
agent.save(path=cfg.model_path)

6
codes/TD3/task1_eval.py
View File

@@ -70,10 +70,10 @@ if __name__ == "__main__":
env.seed(cfg.seed) # Set seeds
torch.manual_seed(cfg.seed)
np.random.seed(cfg.seed)
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.shape[0]
n_states = env.observation_space.shape[0]
n_actions = env.action_space.shape[0]
max_action = float(env.action_space.high[0])
td3= TD3(state_dim,action_dim,max_action,cfg)
td3= TD3(n_states,n_actions,max_action,cfg)
cfg.model_path = './TD3/results/Pendulum-v0/20210428-092059/models/'
cfg.result_path = './TD3/results/Pendulum-v0/20210428-092059/results/'
td3.load(cfg.model_path)

8
codes/TD3/task1_train.py
View File

@@ -79,7 +79,7 @@ def train(cfg,env,agent):
else:
action = (
agent.choose_action(np.array(state))
+ np.random.normal(0, max_action * cfg.expl_noise, size=action_dim)
+ np.random.normal(0, max_action * cfg.expl_noise, size=n_actions)
).clip(-max_action, max_action)
# Perform action
next_state, reward, done, _ = env.step(action)
@@ -109,10 +109,10 @@ if __name__ == "__main__":
env.seed(1) # 随机种子
torch.manual_seed(1)
np.random.seed(1)
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.shape[0]
n_states = env.observation_space.shape[0]
n_actions = env.action_space.shape[0]
max_action = float(env.action_space.high[0])
agent = TD3(state_dim,action_dim,max_action,cfg)
agent = TD3(n_states,n_actions,max_action,cfg)
rewards,ma_rewards = train(cfg,env,agent)
make_dir(plot_cfg.result_path,plot_cfg.model_path)
agent.save(path=plot_cfg.model_path)