Modifying the padding scheme

[joshnunley]

I have an application where I would like to be able to control the padding for the convolution. The way I am doing this now is to just use a much larger matrix than I need, but this is a waste of computation, especially for larger kernels. Is there any simple and efficient way I can use my own padding?

[martinholters]

I'm not completely sure what you'd like to have exactly. Could you elaborate?

[joshnunley]

After thinking about it a bit more, I think what I would like is similar to the option mode='valid' in python (and I think matlab). This mode should be somewhat faster because it doesn't need to compute the convolution with any cells that would need padding.

[martinholters]

IIUC, that means chopping off mix(M,N)-1 elements at both sides of the result as it is computed now, where M and N are the input lengths, with the obvious extension to more than one dimension. Sounds reasonable.