Why the input data should also be warpped by Variable?

I konw the parameters of neural network should be wrapped by Variable to update the value, but the input data never change, why it should also be warpped by Variable?

In the next release Variable will be merged with Tensor, since the Variable class adds indeed some conceptual overhead, which might confuse new PyTorch users.

Here is the GitHub issue, in case you would like to follow the development.

1 Like