AttributeError: 'numpy.ndarray' object has no attribute 'dim' from torch/nn/functional.py

reinforcement-learning
1

I’m working on a DDPG implementation and getting AttributeError: 'numpy.ndarray' object has no attribute 'dim' from my Actor class.
Based on answer here https://discuss.pytorch.org/t/attributeerror-numpy-ndarray-object-has-no-attribute-dim/16026/2, I tried using Variable(), however, the args are Tensors already and I get the TypeError: expected np.ndarray (got Tensor) error.

Code

    def forward(self, state):
        """
        Build the policy/actor network that maps states to actions.
        Using Tanh activation as per DDPG paper
        """
        # print(type(state)) # <class 'torch.Tensor'>
#         state = Variable(T.from_numpy(state))
        x = self.inp(state)
        l1_states = F.relu(x)
        l2_states = F.relu(self.h1(l1_states))
        l3_actions = T.tanh(self.out(l2_states))

        return l3_actions

Error:

<ipython-input-11-d3087ef19a20> in forward(self, state)
     71         # print(type(state)) # <class 'torch.Tensor'>
     72 #         state = Variable(T.from_numpy(state))
---> 73         x = self.inp(state)
     74         l1_states = F.relu(x)
     75         l2_states = F.relu(self.h1(l1_states))

~/anaconda3/envs/drlnd/lib/python3.6/site-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    539             result = self._slow_forward(*input, **kwargs)
    540         else:
--> 541             result = self.forward(*input, **kwargs)
    542         for hook in self._forward_hooks.values():
    543             hook_result = hook(self, input, result)

~/anaconda3/envs/drlnd/lib/python3.6/site-packages/torch/nn/modules/linear.py in forward(self, input)
     85 
     86     def forward(self, input):
---> 87         return F.linear(input, self.weight, self.bias)
     88 
     89     def extra_repr(self):

~/anaconda3/envs/drlnd/lib/python3.6/site-packages/torch/nn/functional.py in linear(input, weight, bias)
   1366         - Output: :math:`(N, *, out\_features)`
   1367     """
-> 1368     if input.dim() == 2 and bias is not None:
   1369         # fused op is marginally faster
   1370         ret = torch.addmm(bias, input, weight.t())

AttributeError: 'numpy.ndarray' object has no attribute 'dim'
2

Based on the error message it looks like state is a numpy array.
Could you check the type again, please, and make sure it’s a tensor?

Don’t use Variables, as they are deprecated since PyTorch 0.4.0.

3

<class 'torch.Tensor'> was the output of print(type(state))

4

Is this error raised directly after the print statement and did you check all state inputs?
Could some of them be numpy arrays somehow?

5

Thanks for the suggestion to check all state inputs. Looks like a bug in my code when I sample from ReplayBuffer. :exploding_head:
I was able to move forward now :+1:

6

Great to hear it’s working! :slight_smile: