def __init__(self, module, name=None):
super(SplitBias, self).__init__()
self.module = module
self.add_bias = AddBias(module.bias.data)
self.module.bias = None
self.name = name
self.module.split_bias = True
def __init__(self,
num_inputs,
num_actions,
recurrent=False,
hidden_size=64):
super(MLPBase_av, self).__init__(recurrent, num_inputs, hidden_size)
if recurrent:
raise NotImplementedError
num_inputs = hidden_size
init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init.
constant_(x, 0), nn.init.calculate_gain('relu'))
self.actor1 = nn.Sequential(init_(nn.Linear(num_inputs, hidden_size)),
nn.Tanh(),
init_(nn.Linear(hidden_size, hidden_size)),
nn.Tanh())
self.critic1 = nn.Sequential(
init_(nn.Linear(num_inputs, hidden_size)), nn.Tanh(),
init_(nn.Linear(hidden_size, hidden_size)), nn.Tanh())
self.actor2 = nn.Sequential(init_(nn.Linear(num_inputs, hidden_size)),
nn.Tanh(),
init_(nn.Linear(hidden_size, hidden_size)),
nn.Tanh())
self.critic2 = nn.Sequential(
init_(nn.Linear(num_inputs, hidden_size)), nn.Tanh(),
init_(nn.Linear(hidden_size, hidden_size)), nn.Tanh())
self.critic_linear1 = init_(nn.Linear(hidden_size, 1))
self.critic_linear2 = init_(nn.Linear(hidden_size, 1))
# Action distribution
init_dist_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init.
constant_(x, 0))
self.fc_mean1 = init_dist_(nn.Linear(hidden_size, num_actions))
self.logstd1 = AddBias(torch.zeros(num_actions))
self.fc_mean2 = init_dist_(nn.Linear(hidden_size, num_actions))
self.logstd2 = AddBias(torch.zeros(num_actions))
self.train()
def __init__(self, num_inputs, num_outputs):
super(DiagGaussian, self).__init__()
init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init.
constant_(x, 0))
self.fc_mean = init_(nn.Linear(num_inputs, num_outputs))
self.logstd = AddBias(torch.zeros(num_outputs))
def __init__(self, num_inputs, num_outputs, init_log_std=0):
super(DiagGaussian, self).__init__()
#init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init.
# constant_(x, 0))
self.fc_mean = nn.Linear(num_inputs, num_outputs)
self.tanh = nn.Tanh()
self.logstd = AddBias(torch.zeros(num_outputs) + init_log_std)
def __init__(self,
obs_shape,
action_space,
base=None,
base_kwargs=None,
dimh=2):
super(Policy, self).__init__()
net = None
dim_input = obs_shape[0]
if len(obs_shape) == 1:
net = [
# actor base
SpLinear(dim_input, dimh, actv_fn='tanh'),
SpLinear(dimh, dimh, actv_fn='tanh'),
# critic base
SpLinear(dim_input, dimh, actv_fn='tanh'),
SpLinear(dimh, dimh, actv_fn='tanh'),
SpLinear(dimh, 1, actv_fn='none'),
]
else:
raise NotImplementedError
if action_space.__class__.__name__ == "Discrete":
self.action_type = 'discrete'
num_outputs = action_space.n
net.append(SpLinear(dimh, num_outputs, actv_fn='none',
init_type=1))
elif action_space.__class__.__name__ == "Box":
self.action_type = 'continuous'
num_outputs = action_space.shape[0]
net.append(SpLinear(dimh, num_outputs, actv_fn='none',
init_type=2))
self.logstd = AddBias(torch.zeros(num_outputs))
else:
raise NotImplementedError
self.net = nn.ModuleList(net)
self.next_layer = {
0: [1],
1: [5],
2: [3],
3: [4],
}
'''
self.next_layer = {
0: [3],
1: [2],
}
'''
self.layers_to_split = list(self.next_layer.keys())
self.n_elites = 64
def __init__(self, num_inputs, num_outputs, zll=False):
super(DiagGaussian, self).__init__()
init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init.
constant_(x, 0))
init_zeros = lambda m: init(
m, lambda x, **kwargs: nn.init.constant_(x, 0), lambda x: nn.init.
constant_(x, 0))
init_last_layer = init_zeros if zll else init_
self.fc_mean = init_last_layer(nn.Linear(num_inputs, num_outputs))
self.logstd = AddBias(torch.zeros(num_outputs))
def __init__(self, num_inputs, num_outputs, activation=None):
super(DiagGaussian, self).__init__()
init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init.
constant_(x, 0))
if activation is None:
self.fc_mean = init_(nn.Linear(num_inputs, num_outputs))
elif activation == "tanh":
# print("!!!")
self.fc_mean = nn.Sequential(
init_(nn.Linear(num_inputs, num_outputs)), nn.Tanh())
else:
raise NotImplementedError
self.logstd = AddBias(torch.zeros(num_outputs))
def __init__(self, module):
super(SplitBias, self).__init__()
self.module = module
self.add_bias = AddBias(module.bias.data)
self.module.bias = None
def __init__(self, num_outputs):
super(DiagGaussianDist, self).__init__()
self.logstd = AddBias(torch.zeros(num_outputs))
