Open tingyuansen opened 9 years ago
The fit to the continuum using sines and cosines is a Gaussian process, actually!
I agree that continuum using sines and cosines is equivalent to setting the 2-points correlation in Gaussian process to be some periodic functions (e.g. sin(a(x-y))). In fact, most, if not all, regressions are some sort of Gaussian process (for e.g. the Cannon is a Gaussian process with 2-points correlation = a_x_y). I am just thinking a periodic Gaussian process is surely less robust than something like square exponential. It seems pretty painless to do that in square exponential (probably less painful than the current fit). Sorry for not being clearer.
Is there any plan using Gaussian processes to do the continuum determination?