Customize Kernels for Conv1d

I would like to extend/customize the Kernel in a Conv1d layer. Is there any example of how this can be done?

Use the functional form torch.nn.functional.conv1d.

1 Like

I don’t think that answers my question, I would like to write my own function: g(t - n) in
(f * g) (t) = sum_n f(n)g(t - n)

Sorry if I’m not clear.

Isn’t that just computing the kernel values, i.e., weights passed to the functional conv1d?