Regarding Conv1d

jit
1

Hi I couldn’t grammatically quite understand the following example

m = nn.Conv1d(16, 33, 3, stride=2)
input = torch.randn(20, 16, 50)
output = m(input)

I think the type of nn.Conv1d is class, and accordingly m should be an instance of the class. Why in the third line, m(input) is used like a function?

Actually, it seems that >>> output = m(input) executes the function forward(self, input) defined in the class nn.Conv1d. So, to be grammatically correct, I think >>> output = m(input) should be changed into >>> output = m.forward(input) . But >>> output = m(input) indeed works.

Could anybody explain a bit about this? What grammatical rule does it follow?

The source code for Conv1d is at the following URL
https://pytorch.org/docs/master/_modules/torch/nn/modules/conv.html#Conv1d

Thank you very much!

2

nn.Module (which nn.Conv1d and all the nn. modules extend), implements the __call__ magic method so that it can be used like a function as in your example. The implementation can be found here. Using __call__ has some extra processing for forward/backward hooks, but, as you have found, its main job is to call forward()

3

Thank you so much for your explanation! Truly helpful!