4

Lets assume I have a some noisy data d(x), like a function f(x) with some noise g(x) which is strongly dependent on x, with d(x) = f(x) + g(x). Is there a clean way to fit or use a spline, lets call it s(x), such that

  • s(x) is smooth
  • s(x) > d(x) for almost all x
  • |s(x) - d(x)| < constant

My idea was to use some filter like Savitzky-Golay to smooth the data and then fit only the maximal part of the result, but actually this method is not very successful.

Cleb
  • 25,102
  • 20
  • 116
  • 151
varantir
  • 6,624
  • 6
  • 36
  • 57
  • My advice would be to model f and g separately and then construct s from f + (upper tail quantile of g). You'll probably get more traction for this question on stats.stackexchange.com or physics.stackexchange.com. – Robert Dodier Oct 11 '14 at 17:19
  • 1
    Could you give examples for f and g, please? – Cleb Jul 15 '15 at 14:09

0 Answers0