def __init__(self, vector_dim, action_dim, quantiles_idx, hidden_units):
super().__init__()
self.action_dim = action_dim
self.q_net_head = mlp(hidden_units['q_net'], out_layer=False) # [B, vector_dim]
self.quantile_net = mlp(hidden_units['quantile'], out_layer=False) # [N*B, quantiles_idx]
self.q_net_tile = mlp(hidden_units['tile'], output_shape=action_dim, out_activation=None) # [N*B, hidden_units['quantile'][-1]]
self(I(shape=vector_dim), I(shape=quantiles_idx))
def __init__(self, vector_dim, action_dim, atoms, hidden_units):
super().__init__()
self.action_dim = action_dim
self.atoms = atoms
self.share = mlp(hidden_units['share'], layer=Noisy, out_layer=False)
self.v = mlp(hidden_units['v'], layer=Noisy, output_shape=atoms, out_activation=None)
self.adv = mlp(hidden_units['adv'], layer=Noisy, output_shape=action_dim * atoms, out_activation=None)
self(I(shape=vector_dim))
def __init__(self, vector_dim, action_dim, options_num, hidden_units, is_continuous=True):
super().__init__()
self.actions_num = action_dim
self.options_num = options_num
self.share = mlp(hidden_units['share'], out_layer=False)
self.q = mlp(hidden_units['q'], output_shape=options_num, out_activation=None)
self.pi = mlp(hidden_units['intra_option'], output_shape=options_num * action_dim, out_activation='tanh' if is_continuous else None)
self.beta = mlp(hidden_units['termination'], output_shape=options_num, out_activation='sigmoid')
self(I(shape=vector_dim))
def __init__(self, vector_dim, output_shape, network_settings):
super().__init__()
self.share = mlp(network_settings['share'], out_layer=False)
self.mu = mlp(network_settings['mu'],
output_shape=output_shape,
out_activation=None)
self.log_std = mlp(network_settings['log_std'],
output_shape=output_shape,
out_activation='tanh')
self(I(shape=vector_dim))
def __init__(self, vector_dim, action_dim, network_settings):
assert len(
network_settings
) > 1, "if you want to use this architecture of critic network, the number of layers must greater than 1"
super().__init__()
self.feature_net = mlp(network_settings[0:1])
self.net = mlp(network_settings[1:],
output_shape=1,
out_activation=None)
self(I(shape=vector_dim), I(shape=action_dim))
def __init__(self, vector_dim, output_shape, network_settings):
super().__init__()
self.share = mlp(network_settings['share'], out_layer=False)
self.logits = mlp(network_settings['logits'],
output_shape=output_shape,
out_activation=None)
self.v = mlp(network_settings['v'],
output_shape=1,
out_activation=None)
self(I(shape=vector_dim))
def __init__(self, vector_dim, output_shape, network_settings):
super().__init__()
self.soft_clip = network_settings['soft_clip']
self.log_std_min, self.log_std_max = network_settings['log_std_bound']
self.share = mlp(network_settings['share'], out_layer=False)
self.mu = mlp(network_settings['mu'],
output_shape=output_shape,
out_activation=None)
self.log_std = mlp(network_settings['log_std'],
output_shape=output_shape,
out_activation=None)
self(I(shape=vector_dim))
def __init__(self, vector_dim, output_shape, head_num, network_settings):
super().__init__()
self.nets = [
mlp(network_settings,
output_shape=output_shape,
out_activation=None) for _ in range(head_num)
]
self(I(shape=vector_dim))
def __init__(self, vector_dim, action_dim, atoms, network_settings):
super().__init__()
self.action_dim = action_dim
self.atoms = atoms
self.net = mlp(network_settings,
output_shape=atoms * action_dim,
out_activation='softmax')
self(I(shape=vector_dim))
def __init__(self, vector_dim, action_dim, nums, network_settings):
super().__init__()
self.action_dim = action_dim
self.nums = nums
self.net = mlp(network_settings,
output_shape=nums * action_dim,
out_activation=None)
self(I(shape=vector_dim))
def __init__(self, vector_dim, output_shape, condition_sigma,
network_settings):
super().__init__()
self.condition_sigma = condition_sigma
self.log_std_min, self.log_std_max = network_settings['log_std_bound']
self.share = mlp(network_settings['hidden_units'], out_layer=False)
self.mu = mlp([], output_shape=output_shape, out_activation='tanh')
if self.condition_sigma:
self.log_std = mlp([],
output_shape=output_shape,
out_activation=None)
else:
self.log_std = tf.Variable(
initial_value=-0.5 *
tf.ones(output_shape, dtype=tf.dtypes.float32),
trainable=True)
self(I(shape=vector_dim))
def __init__(self, vector_dim, output_shape, hidden_units, is_continuous):
super().__init__()
self.is_continuous = is_continuous
out_activation = 'tanh' if self.is_continuous else None
self.net = mlp(hidden_units, act_fn='tanh', output_shape=output_shape, out_activation=out_activation, out_layer=True)
self.weights_2dim = [[i, j] for i, j in zip([vector_dim] + hidden_units, hidden_units + [output_shape])]
self.weights_nums = np.asarray(self.weights_2dim).prod(axis=-1).tolist()
self.weights_total_nums = np.asarray(self.weights_2dim).prod(axis=-1).sum() + np.asarray(hidden_units).sum() + output_shape
self(tf.keras.Input(shape=vector_dim)) # 初始化网络权重
def __init__(self, vector_dim, action_dim, atoms, network_settings):
super().__init__()
self.action_dim = action_dim
self.atoms = atoms
self.net = mlp(network_settings, out_layer=False)
self.outputs = []
for _ in range(action_dim):
self.outputs.append(Dense(atoms, activation='softmax'))
self(I(shape=vector_dim))
def __init__(self,
vector_dim,
output_shape,
network_settings,
out_activation='tanh'):
super().__init__()
self.net = mlp(network_settings,
output_shape=output_shape,
out_activation=out_activation)
self(I(shape=vector_dim))
def __init__(self,
vector_dim,
output_shape,
options_num,
network_settings,
out_activation=None):
super().__init__()
self.actions_num = output_shape
self.options_num = options_num
self.pi = mlp(network_settings,
output_shape=options_num * output_shape,
out_activation=out_activation)
self(I(shape=vector_dim))
def __init__(self,
vector_dim,
action_dim,
options_num,
network_settings,
is_continuous=True):
super().__init__()
self.actions_num = action_dim
self.options_num = options_num
self.share = mlp(network_settings['share'], out_layer=False)
self.q = mlp(network_settings['q'],
output_shape=options_num,
out_activation=None)
self.pi = mlp(network_settings['intra_option'],
output_shape=options_num * action_dim,
out_activation='tanh' if is_continuous else None)
self.beta = mlp(network_settings['termination'],
output_shape=options_num,
out_activation='sigmoid')
self.o = mlp(network_settings['o'],
output_shape=options_num,
out_activation=tf.nn.log_softmax)
self(I(shape=vector_dim))
def __init__(self, vector_dim, output_shape, hidden_units):
super().__init__()
self.logits = mlp(hidden_units, output_shape=output_shape, out_activation=None)
self(I(shape=vector_dim))
def __init__(self, vector_dim, action_dim, hidden_units):
super().__init__()
self.net = mlp(hidden_units, output_shape=1, out_activation=None)
self(I(shape=vector_dim), I(shape=action_dim))
def __init__(self, vector_dim, output_shape, head_num, hidden_units):
super().__init__()
self.nets = [mlp(hidden_units, output_shape=output_shape, out_activation=None) for _ in range(head_num)]
self(I(shape=vector_dim))
def __init__(self, vector_dim, output_shape, options_num, hidden_units, out_activation=None):
super().__init__()
self.actions_num = output_shape
self.options_num = options_num
self.pi = mlp(hidden_units, output_shape=options_num * output_shape, out_activation=out_activation)
self(I(shape=vector_dim))
def __init__(self, vector_dim, output_shape, hidden_units):
super().__init__()
self.share = mlp(hidden_units['share'], out_layer=False)
self.logits = mlp(hidden_units['logits'], output_shape=output_shape, out_activation=None)
self.v = mlp(hidden_units['v'], output_shape=1, out_activation=None)
self(I(shape=vector_dim))
def __init__(self, vector_dim, network_settings):
super().__init__()
self.net = mlp(network_settings, output_shape=1, out_activation=None)
self(I(shape=vector_dim))
def __init__(self, vector_dim, action_dim, atoms, hidden_units):
super().__init__()
self.action_dim = action_dim
self.atoms = atoms
self.net = mlp(hidden_units, output_shape=atoms * action_dim, out_activation='softmax')
self(I(shape=vector_dim))
def __init__(self, vector_dim, action_dim, nums, hidden_units):
super().__init__()
self.action_dim = action_dim
self.nums = nums
self.net = mlp(hidden_units, output_shape=nums * action_dim, out_activation=None)
self(I(shape=vector_dim))
def __init__(self, vector_dim, output_shape, hidden_units):
super().__init__()
self.net = mlp(hidden_units, output_shape=output_shape, out_activation='tanh')
self(I(shape=vector_dim))
