def __init__(self,
observation_space,
action_space,
net,
rew_func,
n_samples=1000,
horizon=20,
mean_obs=0.,
std_obs=1.,
mean_acs=0.,
std_acs=1.,
rnn=False,
normalize_ac=True):
BasePol.__init__(self,
observation_space,
action_space,
net,
rnn=rnn,
normalize_ac=normalize_ac)
self.rew_func = rew_func
self.n_samples = n_samples
self.horizon = horizon
self.to(get_device())
self.mean_obs = torch.tensor(mean_obs,
dtype=torch.float).repeat(n_samples, 1)
self.std_obs = torch.tensor(std_obs,
dtype=torch.float).repeat(n_samples, 1)
self.mean_acs = torch.tensor(mean_acs,
dtype=torch.float).repeat(n_samples, 1)
self.std_acs = torch.tensor(std_acs,
dtype=torch.float).repeat(n_samples, 1)
def __init__(self, ob_space, ac_space, qfunc, rnn=False, normalize_ac=True, data_parallel=False, parallel_dim=0, eps=0.2):
BasePol.__init__(self, ob_space, ac_space, None, rnn,
normalize_ac, data_parallel, parallel_dim)
self.qfunc = qfunc
self.eps = eps
self.a_i_shape = (1, )
self.to(get_device())
def __init__(self,
observation_space,
action_space,
net,
normalize_ac=True):
BasePol.__init__(self, observation_space, action_space, normalize_ac)
self.net = net
self.pd = MixtureGaussianPd()
self.to(get_device())
def __init__(self,
observation_space,
action_space,
net,
rnn=False,
normalize_ac=True):
BasePol.__init__(self, observation_space, action_space, net, rnn,
normalize_ac)
self.pd = GaussianPd()
self.to(get_device())
def __init__(self,
observation_space,
action_space,
net,
rnn=False,
normalize_ac=True):
BasePol.__init__(self, observation_space, action_space, net, rnn,
normalize_ac)
self.pd = MultiCategoricalPd()
self.to(get_device())
def __init__(self,
observation_space,
action_space,
net,
rnn=False,
normalize_ac=True,
data_parallel=False,
parallel_dim=0):
BasePol.__init__(self, observation_space, action_space, net, rnn,
normalize_ac, data_parallel, parallel_dim)
self.pd = GaussianPd()
self.to(get_device())
def __init__(self,
ob_space,
ac_space,
net,
rnn=False,
normalize_ac=True,
data_parallel=False,
parallel_dim=0):
BasePol.__init__(self, ob_space, ac_space, net, rnn, normalize_ac,
data_parallel, parallel_dim)
self.pd = MultiCategoricalPd()
self.to(get_device())
def __init__(self,
observation_space,
action_space,
qfunc,
rnn=False,
normalize_ac=True,
eps=0.2):
BasePol.__init__(self, observation_space, action_space, None, rnn,
normalize_ac)
self.qfunc = qfunc
self.eps = eps
self.a_i_shape = (1, )
self.to(get_device())
def __init__(self,
observation_space,
action_space,
net=None,
rnn=False,
normalize_ac=True,
data_parallel=False,
parallel_dim=0):
BasePol.__init__(self,
observation_space,
action_space,
net,
rnn=rnn,
normalize_ac=normalize_ac,
data_parallel=data_parallel,
parallel_dim=parallel_dim)
def __init__(self, ob_space, ac_space, net, rew_func, n_samples=1000, horizon=20,
mean_obs=0., std_obs=1., mean_acs=0., std_acs=1., rnn=False,
normalize_ac=True, data_parallel=False, parallel_dim=0):
BasePol.__init__(self, ob_space, ac_space, net, rnn=rnn, normalize_ac=normalize_ac,
data_parallel=data_parallel, parallel_dim=parallel_dim)
self.rew_func = rew_func
self.n_samples = n_samples
self.horizon = horizon
self.to(get_device())
self.mean_obs = torch.tensor(
mean_obs, dtype=torch.float).repeat(n_samples, 1)
self.std_obs = torch.tensor(
std_obs, dtype=torch.float).repeat(n_samples, 1)
self.mean_acs = torch.tensor(
mean_acs, dtype=torch.float).repeat(n_samples, 1)
self.std_acs = torch.tensor(
std_acs, dtype=torch.float).repeat(n_samples, 1)
def __init__(self,
observation_space,
action_space,
net,
noise=None,
rnn=False,
normalize_ac=True):
if rnn:
raise ValueError(
'rnn with DeterministicActionNoisePol is not supported now')
BasePol.__init__(self,
observation_space,
action_space,
net,
rnn=rnn,
normalize_ac=normalize_ac)
self.noise = noise
self.pd = DeterministicPd()
self.to(get_device())
def __init__(self,
ob_space,
ac_space,
net,
noise=None,
rnn=False,
normalize_ac=True,
data_parallel=False,
parallel_dim=0):
if rnn:
raise ValueError(
'rnn with DeterministicActionNoisePol is not supported now')
BasePol.__init__(self,
ob_space,
ac_space,
net,
rnn=rnn,
normalize_ac=normalize_ac,
data_parallel=data_parallel,
parallel_dim=parallel_dim)
self.noise = noise
self.pd = DeterministicPd()
self.to(get_device())
def __init__(self, observation_space, action_space, net=None, rnn=False, normalize_ac=True):
BasePol.__init__(self, observation_space, action_space, net, rnn=rnn,
normalize_ac=normalize_ac)
