On Sun, 22 Aug 2010, Andy Farnell wrote:
On Sun, 22 Aug 2010 15:38:50 +0200 Ludwig Maes ludwig.maes@gmail.com wrote:
probably some offset problem? or hardcoded assumption that it will be longer than 2?
Convolution doesn't really make much sense for 1 where it degenerates to simple multiplication.
Multiplication by 1 doesn't make much sense, where it degenerates to not multiplying. Yet it's allowed. Why ?
To unify with the special case of "here we don't really have to multiply". Because it's less complicated like that and it doesn't cost anything to allow this feature.
as an example of this uniformity principle, GridFlow's [#convolve] allow convolution kernels of size 1 and of size 0, just because there's no real reason to forbid them. In such cases, they do what is expected of them : multiply by a constant, and multiply by zero, respectively.
And then, if Ludwig said it right, [FIR~] wouldn't work for a kernel size of 2... what would you say ? that it doesn't make much sense because you can do that with [rzero~] and [*~] together ?
_ _ __ ___ _____ ________ _____________ _____________________ ... | Mathieu Bouchard, Montréal, Québec. téléphone: +1.514.383.3801