def dueling_conv_net(env, frames=4):
return nn.Sequential(
nn.Conv2d(frames, 32, 8, stride=4),
nn.ReLU(),
nn.Conv2d(32, 64, 4, stride=2),
nn.ReLU(),
nn.Conv2d(64, 64, 3, stride=1),
nn.ReLU(),
Flatten(),
Dueling(
nn.Sequential(
nn.Linear(3456, 512),
nn.ReLU(),
Linear0(512, 1)
),
nn.Sequential(
nn.Linear(3456, 512),
nn.ReLU(),
Linear0(512, env.action_space.n)
),
)
)
def policy_net(env):
return nn.Sequential(
nn.Linear(3456, 512), nn.ReLU(), Linear0(512, env.action_space.n)
)
def test_linear0(self):
model = Linear0(3, 3)
result = model(torch.tensor([[3., -2., 10]]))
tt.assert_equal(result, torch.tensor([[0., 0., 0.]]))
def value_net():
return nn.Sequential(nn.Linear(3456, 512), nn.ReLU(), Linear0(512, 1))