I want to smooth peaks for non-negative matrix factorization (NMF). Since Savitzky-Golay has polynomial smoothing, I think it is good for narrow peak shape in my signal profile. But NMF needs to be non-negative, I wonder I can simply take absolute on Savtizky-Golay filter (or replace negative value with eps (1e-6))? Are there any published applications using only positive range of Savitzky-Golay to smooth data? Or are there other appropriate filter for non-negative data?

  • 1
    $\begingroup$ I think your primary question is "Is Savitzky-Golay appropriate for my problem?"; I think that would make a good question on here, together with an explanation of what your problem is and why you're considering a modification of SG. Without having put too much thought into it: no, if you just clip at negative values, you do horrible things to your signal and its spectrum, and unless you have a really funky signal model that doesn't care about these acts of signal terrorism, I'd advice you to probably not do that. So, ask that "is SG the right approach?" question in a new post; I'd read that. $\endgroup$ Commented Nov 27, 2018 at 11:41


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.