forum.alglib.net

I am new to alglib and non-linear smoothing methods. looking at the attached graphic, i want to reduce the jaggies on these channel boundary pairs to a smooth appearance while preserving the over all shape of the top and bottom of each of the 3 channel pairs shown. I would be willing to hire someone to create a c++ example. thanks... thinkpad@edgemax.com

I recommend you to wait for ALGLIB 3.1, which will be released somewhere between October, 31, and November, 1. It supports penalized regression splines, which are perfectly suited for problems like you want to solve. They allow to adaptively (and continuously) control degree of smoothing applied to data: from noisy graphs above to smooth shape-preserving fit to straight line fit. Unpenalized regression splines (which ALGLIB supports now) are inferior to the penalized ones.

BTW, how large are the samples you want to process?

P.S. Forgot to tell that ALGLIB 3.1 will include examples for penalized regression splines =)

Thanks, this sounds very promising! One question regards Penalized Regression Splines. Are PRS computationally expensive? Ideally, I would like to update the smoothing spline output as data comes in from the sensors. Perhaps using a sliding window perhaps 50 samples wide. The data rate is sporadic and not constant, sometimes several times per second.

They have O(N*M^2 + M^3) complexity, where N is a number of points (50 in your case), M is a number of basis functions (recommended value for your problem is 100-200 - several times more than number of points). Generally, it takes approximately same time as to solve MxM system of linear equations, which should be acceptable if you use 2GHz CPU or higher. In case it is something embedded (=slow), you can decrease number of basis functions - or wait for ALGLIB 3.2, which will contain faster algorithm for spline fitting with O(N*M+M*M) complexity :)

What is the difference between smoothing Spline and moving averge and is alglib 3.4 has Moving average filter