def __init__(self, num_inputs, recurrent=False, num_streets=4, hidden_size=256, total_hidden_size=512): if recurrent: raise NotImplementedError("recurrent policy not done yet") super(NaviBaseTemp, self).__init__(recurrent, hidden_size, hidden_size) self.num_streets = num_streets init_cnn = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init. constant_(x, 0), nn.init.calculate_gain('relu')) self.img_embed = nn.Sequential( init_cnn(nn.Conv2d(8, 96, 3, stride=2)), nn.ReLU(), init_cnn(nn.Conv2d(96, 96, 5, stride=2)), nn.ReLU(), init_cnn(nn.Conv2d(96, 32, 5, stride=2)), nn.ReLU(), Flatten(), init_cnn(nn.Linear(32 * 8 * 8, total_hidden_size)), nn.ReLU()) init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init. constant_(x, 0)) self.critic_linear = init_(nn.Linear(total_hidden_size, 1)) self.train()
def __init__(self, num_inputs, num_outputs): super(Bernoulli, self).__init__() init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init. constant_(x, 0)) self.linear = init_(nn.Linear(num_inputs, num_outputs))
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, recurrent=False, num_streets=4, hidden_size=256, total_hidden_size=(256 * 10)): if recurrent: raise NotImplementedError("recurrent policy not done yet") super(NaviBase, self).__init__(recurrent, hidden_size, hidden_size) self.num_streets = num_streets init_cnn = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init. constant_(x, 0), nn.init.calculate_gain('relu')) init_dense = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init. constant_(x, 0), np.sqrt(2)) self.img_embed = nn.Sequential( init_cnn(nn.Conv2d(3, 32, 3, stride=2)), nn.ReLU(), init_cnn(nn.Conv2d(32, 64, 5, stride=2)), nn.ReLU(), init_cnn(nn.Conv2d(64, 32, 5, stride=2)), nn.ReLU(), Flatten(), init_cnn(nn.Linear(32 * 8 * 8, hidden_size)), nn.ReLU()) # NeED to look if different activation functions self.coord_embed = nn.Sequential( init_dense(nn.Linear(2, 64)), nn.Tanh(), init_dense(nn.Linear(64, hidden_size)), nn.Tanh()) self.number_embed = nn.Sequential(init_dense(nn.Linear(10, 64)), nn.Tanh()) self.street_embed = nn.Sequential( init_dense(nn.Linear(self.num_streets, hidden_size)), nn.Tanh()) init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init. constant_(x, 0)) self.fusion = nn.Sequential( init_dense(nn.Linear(total_hidden_size, total_hidden_size)), nn.Tanh(), init_dense(nn.Linear(total_hidden_size, total_hidden_size)), nn.Tanh()) self.critic_linear = init_(nn.Linear(total_hidden_size, 1)) self.train()
def __init__(self, num_inputs, num_outputs): super(Categorical, self).__init__() init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init.constant_(x, 0), gain=0.01) self.linear = init_(nn.Linear(num_inputs, num_outputs))
def __init__(self, num_inputs, recurrent=False, hidden_size=512): super(CNNBase, self).__init__(recurrent, hidden_size, hidden_size) init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init. constant_(x, 0), nn.init.calculate_gain('relu')) self.main = nn.Sequential( init_(nn.Conv2d(num_inputs, 32, 8, stride=4)), nn.ReLU(), init_(nn.Conv2d(32, 64, 4, stride=2)), nn.ReLU(), init_(nn.Conv2d(64, 32, 3, stride=1)), nn.ReLU(), Flatten(), init_(nn.Linear(32 * 7 * 7, hidden_size)), nn.ReLU()) init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init. constant_(x, 0)) self.critic_linear = init_(nn.Linear(hidden_size, 1)) self.train()
def __init__(self, num_inputs, recurrent=False, hidden_size=64): super(MLPBase, self).__init__(recurrent, num_inputs, hidden_size) if recurrent: num_inputs = hidden_size init_ = lambda m: init(m, nn.init.orthogonal_, lambda x: nn.init. constant_(x, 0), np.sqrt(2)) self.actor = nn.Sequential(init_(nn.Linear(num_inputs, hidden_size)), nn.Tanh(), init_(nn.Linear(hidden_size, hidden_size)), nn.Tanh()) self.critic = nn.Sequential(init_(nn.Linear(num_inputs, hidden_size)), nn.Tanh(), init_(nn.Linear(hidden_size, hidden_size)), nn.Tanh()) self.critic_linear = init_(nn.Linear(hidden_size, 1)) self.train()