def __init__(self, env, is_batch_norm=False, is_grad_inverter=True):
super().__init__(env)
assert isinstance(env.action_space,
Box), "action space must be continuous"
if is_batch_norm:
self.critic_net = CriticNet_bn(self.observation_space_size,
self.action_space_size)
self.actor_net = ActorNet_bn(self.observation_space_size,
self.action_space_size)
else:
self.critic_net = CriticNet(self.observation_space_size,
self.action_space_size)
self.actor_net = ActorNet(self.observation_space_size,
self.action_space_size)
self.is_grad_inverter = is_grad_inverter
self.replay_memory = deque()
self.time_step = 0
action_max = np.array(self.high).tolist()
action_min = np.array(self.low).tolist()
action_bounds = [action_max, action_min]
self.grad_inv = grad_inverter(action_bounds)
def __init__(self,env, is_batch_norm=False):
self.env = env
self.num_states = env.observation_space.shape[0]
self.num_actions = env.action_space.shape[0]
if is_batch_norm:
self.critic_net = CriticNet_bn(self.num_states, self.num_actions)
self.actor_net = ActorNet_bn(self.num_states, self.num_actions)
else:
self.critic_net = CriticNet(self.num_states, self.num_actions)
self.actor_net = ActorNet(self.num_states, self.num_actions)
#Initialize Buffer Network:
self.replay_memory = deque()
#Intialize time step:
self.time_step = 0
self.counter = 0
action_max = np.array(env.action_space.high).tolist()
action_min = np.array(env.action_space.low).tolist()
action_bounds = [action_max,action_min]
self.grad_inv = grad_inverter(action_bounds)
def __init__(self,env, is_batch_norm):
self.env = env
self.num_states = 59
self.num_actions = 3
if is_batch_norm:
self.critic_net = CriticNet_bn(self.num_states, self.num_actions)
self.actor_net = ActorNet_bn(self.num_states, self.num_actions)
else:
self.critic_net = CriticNet(self.num_states, self.num_actions)
self.actor_net = ActorNet(self.num_states, self.num_actions)
#Initialize Buffer Network:
self.replay_memory = deque()
#Intialize time step:
self.time_step = 0
self.counter = 0
action_max = [1.0, 1.0, 1.0]
action_min = [-1.0, -1.0, -1.0]
action_bounds = [action_max,action_min]
self.grad_inv = grad_inverter(action_bounds)
def __init__(self, num_states, num_actions, is_batch_norm):
self.num_states = num_states
self.num_actions = num_actions
if is_batch_norm:
self.critic_net = CriticNet_bn(self.num_states, self.num_actions)
self.actor_net = ActorNet_bn(self.num_states, self.num_actions)
else:
self.critic_net = CriticNet(self.num_states, self.num_actions)
self.actor_net = ActorNet(self.num_states, self.num_actions)
#Initialize Buffer Network:
self.replay_memory = deque()
#Intialize time step:
self.time_step = 0
self.counter = 0
action_max = 5 * np.ones((1, num_actions))
action_max = action_max.flatten()
action_max = action_max.tolist()
action_min = 0 * np.ones((1, num_actions))
action_min = action_min.flatten()
action_min = action_min.tolist()
action_bounds = [action_max, action_min]
self.grad_inv = grad_inverter(action_bounds)
def __init__( self, hisar_size, ar_size, action_size, TAU = 0.001, is_batch_norm = 0, write_sum = 0, net_size_scale=1, max_load=1, beta0=beta):
self.hisar_size = hisar_size
self.load_size = action_size + 1
self.ar_size = ar_size
self.state_size = action_size * 2
self.action_size = action_size
self.ar_action_size = ar_size + action_size
#print("net_size_scale: "+str(net_size_scale))
if is_batch_norm:
if len(CN_N_HIDDENS)==2:
self.critic_net = CriticNet_bn( self.state_size, self.action_size, TAU, write_sum, net_size_scale )
else:
self.critic_net = CriticNet_bn_3( self.state_size, self.action_size, TAU, write_sum, net_size_scale )
self.actor_net = ActorNet_bn( self.state_size, self.action_size, TAU, write_sum, net_size_scale )
self.ar_pred_net = ARPredNet_bn( self.hisar_size, self.ar_size, write_sum, net_size_scale ) # arrival rate prediction network
self.load_map_net = LoadMapNet_bn( self.ar_size, self.action_size, self.load_size, write_sum, net_size_scale ) # load mapping network
else:
self.critic_net = CriticNet( self.state_size, self.action_size, TAU, write_sum, net_size_scale )
self.actor_net = ActorNet( self.state_size, self.action_size, TAU, write_sum, net_size_scale )
self.ar_pred_net = ARPredNet( self.hisar_size, self.ar_size, write_sum, net_size_scale ) # arrival rate prediction network
self.load_map_net = LoadMapNet( self.ar_size, self.action_size, self.load_size, write_sum, net_size_scale ) # load mapping network
self.env = ENV( action_size, max_load=max_load, beta0=beta0 )
#self.k_nearest_neighbors = int(max_actions * k_ratio )
#Initialize Network Buffers:
self.replay_memory_ac = deque()
self.replay_memory_arp = deque()
self.replay_memory_lm = deque()
#Intialize time step:
self.time_step = 0
self.counter = 0
action_max = np.ones( ( self.action_size ) ).tolist()
action_min = np.zeros( ( self.action_size ) ).tolist()
action_bounds = [action_max, action_min]
self.grad_inv = grad_inverter( action_bounds )