Serving the Quantitative Finance Community
Convoluation algorithm
Hi, All,Anyone knows very fast convoluaton algorithm? I don't want to bother FFT.Thanks,CR
Last edited by caperover on June 10th, 2008, 10:00 pm, edited 1 time in total.
-
- katastrofa
- Posts: 7954
- Joined:
- Location: Event Horizon
Convoluation algorithm
In the general case, I'd use FFT (it's not hard...).When one of the convoluted sequences has mostly zeros, a straightforward calculation will be faster.
Convoluation algorithm
QuoteOriginally posted by: caperoverHi, All,Anyone knows very fast convoluaton algorithm? I don't want to bother FFT.Thanks,CRI'm not a specialist but it seems to me that speed is an issue for you.FFT is what you need. It is, for instance, already implemented in the GSL.What language do you use? What speed do you need? (How large data structures and how many times the calculation?)blondie.
-
- zarnywhoop
- Posts: 0
- Joined:
Convoluation algorithm
QuoteOriginally posted by: caperoverHi, All,Anyone knows very fast convoluaton algorithm? I don't want to bother FFT.Thanks,CRFFT is the very fast algorithm
Convoluation algorithm
FFT is much faster than directly implementing convolution. Never heard that people do convolution without using FFT. FFT in matlab is super-easy as only one command, should not be complex in GSL. Good luck.
Last edited by tonney on June 15th, 2008, 10:00 pm, edited 1 time in total.
-
- katastrofa
- Posts: 7954
- Joined:
- Location: Event Horizon
Convoluation algorithm
In Matlab convolution via FFT is faster. In Java, I've observed that when one distribution had only 2 non-zero elements (and I knew it), it was faster to do it directly.