Different behavior of autograd.Variable

autograd
1

I’m getting two different behaviors of autograd.Variable and I’m wondering if it’s because of the pytorch version.

On my local machine (OS X, pytorch version 0.4.0), I get

In [6]: type(autograd.Variable(torch.from_numpy(np.array([10,2]))))
Out[6]: torch.Tensor

On the server I’m using (pytorch version 0.3.0.post4), I get

In [6]: type(autograd.Variable(torch.from_numpy(np.array([10,2]))))
Out[6]: torch.autograd.variable.Variable

This different behavior affects my code downstream because those two instance types have different methods (e.g., topk() vs. argmax()).

Any ideas?

2

Tensors and Variables were merged in version 0.4.0.
Have a look at the Migration Guide.

3

Thanks, this clears things up!