Why odes F.relu not able to process numpy arrays?

1

I did:

F.relu(np.ones(3))

but it threw an error:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/brandomiranda/miniconda3/envs/hbf_env/lib/python3.6/site-packages/torch/nn/functional.py", line 463, in relu
    return _functions.thnn.Threshold.apply(input, 0, 0, inplace)
  File "/Users/brandomiranda/miniconda3/envs/hbf_env/lib/python3.6/site-packages/torch/nn/_functions/thnn/auto.py", line 126, in forward
    ctx._backend = type2backend[type(input)]
  File "/Users/brandomiranda/miniconda3/envs/hbf_env/lib/python3.6/site-packages/torch/_thnn/__init__.py", line 15, in __getitem__
    return self.backends[name].load()
KeyError: <class 'numpy.ndarray'>

why? do torch ops not process numpy stuff innately?

2

seems this issue also happens for torch.nn.ReLU

3

F.relu operates on a Variables containing a Tensor object and not on a numpy ndarray.

If you want to use torch Functions (such as F.relu), convert your numpy ndarray to a torch Tensor as:

np_arr = np.array([0,1,2,3])
tensor_arr = torch.from_numpy(np_arr)
tensor_var = Variable(tensor_arr)

F.relu(tensor_var)
4

yea but now I need to wrap F.relu and change types and check types etc seems unnecessary, but why would it need to be like this?

5

that’s how the library works. We dont process numpy arrays without conversion into torch Tensors. numpy arrays are not even defined for the GPU.