def __init__(self, args):
super(A3C_LSTM_NN, self).__init__(args)
self.conv1 = nn.Conv2d(3, 32, 5, stride=1, padding=2)
self.down1 = nn.Conv2d(32, 32, 3, stride=2, padding=1)
self.conv2 = nn.Conv2d(32, 32, 5, stride=1, padding=2)
self.down2 = nn.Conv2d(32, 32, 3, stride=2, padding=1)
self.conv3 = nn.Conv2d(32, 64, 3, stride=1, padding=1)
self.down3 = nn.Conv2d(64, 64, 3, stride=2, padding=1)
self.conv4 = nn.Conv2d(64, 64, 3, stride=1, padding=0)
self.down4 = nn.Conv2d(64, 64, 3, stride=2, padding=1)
self.lstm = nn.LSTMCell(1024, self.hidden_dim)
self.linear_encoder = nn.Linear(self.hidden_dim, self.hidden_dim)
self.linear_mu = nn.Linear(self.hidden_dim, self.hidden_dim)
self.bn_x = nn.BatchNorm1d(self.hidden_dim * 2)
self.policy_5 = NoisyLinear(self.hidden_dim * 2, self.output_dims)
self.policy_6 = nn.Softmax(dim=1)
self.value_5 = NoisyLinear(self.hidden_dim * 2, 1)
self.sampler = sampler(args)
self._reset()
self.train()
def __init__(self, args):
super(A3C_LSTM_NN_CRELU, self).__init__(args)
print("USING NEW MODEL with CRELU")
self.conv1 = nn.Conv2d(3, 16, 5, stride=1, padding=2)
self.down1 = nn.Conv2d(32, 32, 3, stride=2, padding=1)
self.crelu1 = crelu()
self.conv2 = nn.Conv2d(32, 16, 5, stride=1, padding=2)
self.down2 = nn.Conv2d(32, 32, 3, stride=2, padding=1)
self.crelu2 = crelu()
self.conv3 = nn.Conv2d(32, 32, 3, stride=1, padding=1)
self.down3 = nn.Conv2d(64, 64, 3, stride=2, padding=1)
self.crelu3 = crelu()
self.conv4 = nn.Conv2d(64, 32, 3, stride=1, padding=0)
self.down4 = nn.Conv2d(64, 64, 3, stride=2, padding=1)
self.crelu4 = crelu()
self.lstm = nn.LSTMCell(1024, self.hidden_dim)
self.linear_encoder = nn.Linear(self.hidden_dim,
int(self.hidden_dim / 2))
self.linear_mu = nn.Linear(self.hidden_dim, int(self.hidden_dim / 2))
self.crelu = crelu()
self.crelu_encoder = crelu()
self.policy_5 = NoisyLinear(self.hidden_dim * 2, self.output_dims)
self.policy_6 = nn.Softmax()
self.value_5 = NoisyLinear(self.hidden_dim * 2, 1)
self.sampler = sampler(args)
self._reset()
self.train()
def __init__(self, args):
super(A3C_LSTM_HPA3C, self).__init__(args)
if args.crelu:
self.crelu = True
else:
self.crelu = False
self.sig = args.sig
self.conv1 = nn.Conv2d(3, 32, 5, stride=1, padding=2)
self.down1 = nn.Conv2d(32, 32, 3, stride=2, padding=1)
self.conv2 = nn.Conv2d(32, 32, 5, stride=1, padding=2)
self.down2 = nn.Conv2d(32, 32, 3, stride=2, padding=1)
self.conv3 = nn.Conv2d(32, 64, 3, stride=1, padding=1)
self.down3 = nn.Conv2d(64, 64, 3, stride=2, padding=1)
self.conv4 = nn.Conv2d(64, 64, 3, stride=1, padding=0)
self.down4 = nn.Conv2d(64, 64, 3, stride=2, padding=1)
self.linear_mu = nn.Linear(self.hidden_dim, self.hidden_dim)
self.linear_sigma = nn.Linear(self.hidden_dim, self.hidden_dim)
self.prior_mu = nn.Linear(self.hidden_dim, self.hidden_dim)
self.lstm = nn.LSTMCell(1024, self.hidden_dim)
self.linear_encoder = nn.Linear(self.hidden_dim, self.hidden_dim)
self.policy_5 = nn.Linear(self.hidden_dim * 2, self.output_dims)
self.policy_6 = nn.Softmax(dim=1)
self.value_5 = nn.Linear(self.hidden_dim * 2, 1)
self.sampler = sampler(args)
if self.crelu:
self.crelu_z = crelu()
self.crelu_x = crelu()
self.prior_mu = nn.Linear(self.hidden_dim * 2, self.hidden_dim)
self.policy_5 = nn.Linear(self.hidden_dim * 4, self.output_dims)
self.value_5 = nn.Linear(self.hidden_dim * 4, 1)
self._reset()
self.train()