def __init__(self,
s_dim,
a_dim,
h_dim,
h_act=nn.ReLU,
buffer_size=100000,
batch_size=32,
lr=1e-4,
gamma=0.95,
theta=0.01,
noise_level=None,
n_sample=5,
*args,
**kwargs):
super(B3DQNAgent, self).__init__()
self.q_net = BayesNet(in_dim=s_dim,
o_dim=a_dim,
h_dim=h_dim,
h_act=h_act,
noise_level=noise_level)
self.target_net = BayesNet(in_dim=s_dim,
o_dim=a_dim,
h_dim=h_dim,
h_act=h_act,
noise_level=noise_level)
self.target_net.load_state_dict(self.q_net.state_dict())
self.buffer = ReplayBuffer(buffer_size)
self.batch_size = batch_size
self.optimizer = Adam(self.q_net.parameters(), lr=lr)
self.gamma = gamma
self.theta = theta
self.noise_level = noise_level
self.n_sample = n_sample
self.a_dim = a_dim
def __init__(self,
s_dim,
a_dim,
h_dim,
h_act=nn.ReLU,
buffer_size=100000,
batch_size=32,
lr=1e-4,
gamma=0.95,
theta=0.01,
*args,
**kwargs):
super(DQNAgent, self).__init__()
self.q_net = SimpleMLP(in_dim=s_dim,
o_dim=a_dim,
h_dim=h_dim,
h_act=h_act)
self.target_net = SimpleMLP(in_dim=s_dim,
o_dim=a_dim,
h_dim=h_dim,
h_act=h_act)
self.target_net.load_state_dict(self.q_net.state_dict())
self.buffer = ReplayBuffer(buffer_size)
self.batch_size = batch_size
self.optimizer = Adam(self.q_net.parameters(), lr=lr)
self.gamma = gamma
self.theta = theta
self.a_dim = a_dim
def __init__(self,
s_dim,
a_dim,
h_dim,
h_act=nn.ReLU,
buffer_size=100000,
batch_size=32,
lr=1e-4,
gamma=0.95,
theta=0.01,
n_model=5,
*args,
**kwargs):
super(BootDQNAgent, self).__init__()
q_list = [
SimpleMLP(in_dim=s_dim, o_dim=a_dim, h_dim=h_dim, h_act=h_act)
for _ in range(n_model)
]
target_list = [
SimpleMLP(in_dim=s_dim, o_dim=a_dim, h_dim=h_dim, h_act=h_act)
for _ in range(n_model)
]
self.q_nets = nn.ModuleList(q_list)
self.target_nets = nn.ModuleList(target_list)
self.target_nets.load_state_dict(self.q_nets.state_dict())
self.buffer = ReplayBuffer(buffer_size)
self.batch_size = batch_size
self.optimizers = [
Adam(q_net.parameters(), lr=lr) for q_net in self.q_nets
]
self.gamma = gamma
self.theta = theta
self.n_model = n_model
self.current_head = None
self.a_dim = a_dim
def __init__(self,
s_dim,
a_dim,
h_dim,
h_act=nn.ReLU,
buffer_size=100000,
batch_size=32,
lr=1e-4,
gamma=0.95,
theta=0.01,
n_model=5,
*args,
**kwargs):
super(EnDQNAgent, self).__init__()
self.q_nets = EnModel(in_dim=s_dim,
o_dim=a_dim,
h_dim=h_dim,
h_act=h_act,
n_model=5)
self.target_nets = EnModel(in_dim=s_dim,
o_dim=a_dim,
h_dim=h_dim,
h_act=h_act,
n_model=5)
self.target_nets.load_state_dict(self.q_nets.state_dict())
self.buffer = ReplayBuffer(buffer_size)
self.batch_size = batch_size
self.optimizers = [
Adam(head.parameters(), lr=lr) for head in self.q_nets.heads
]
self.gamma = gamma
self.theta = theta
self.n_model = n_model
self.a_dim = a_dim
def __init__(self,
s_dim,
a_dim,
h_dim,
h_act=nn.ReLU,
buffer_size=100000,
batch_size=32,
lr=1e-4,
gamma=0.95,
theta=0.01,
dropout=0.5,
weight_decay=0.1,
noise_level=None,
n_sample=5,
*args,
**kwargs):
super(DropDQNAgent, self).__init__()
self.q_net = MCDropout(in_dim=s_dim,
o_dim=a_dim,
h_dim=h_dim,
h_act=h_act,
dropout=dropout,
noise_level=noise_level)
self.target_net = MCDropout(in_dim=s_dim,
o_dim=a_dim,
h_dim=h_dim,
h_act=h_act,
dropout=dropout,
noise_level=noise_level,
agent=False)
self.target_net.load_state_dict(self.q_net.state_dict())
self.buffer = ReplayBuffer(buffer_size)
self.batch_size = batch_size
self.optimizer = Adam(self.q_net.parameters(),
lr=lr,
weight_decay=weight_decay)
self.gamma = gamma
self.theta = theta
self.noise_level = noise_level
self.n_sample = n_sample
self.a_dim = a_dim