forum.alglib.net

I wrote some time ago for help on an approach but I have not got the results I expected. With the function you suggested me to use do not get a really good approximation. Before I try to implement the program with c# I did in MATLAB and approached the curve quite well. I send a picture with what I get with MATLAB and with your spline1dfitcubicwc function. The blue line are the points to align, the red line I get with MATLAB and green which I get from spline1dfitcubicwc function. Thanks in advance.

I forgot ... I do an analysis of the sections which I have and analyze each section separately

Can you describe your problem in more details:
* do you have parametric problem (you try to fit x(t) and y(t) separately) or you try to fit Y as function of X?
* how do you split your data into segments?

XY points are geographical coordinates (lat lon). This point define the trajectory of a mobile. Data analyzed are not very stable as you can see then what I have to do is bring that information to then represent them in google maps and you're in line (more or less) to the reality of the streets. So I want to approach and no interpolations. If you look at the red line I get with MATLAB is more or less the direction of the street but not exactly.

To split data into segments I had create a function who analyze the tendence of the curve and after, cut up the curve in segments

Do you have time values assigned to points? I.e. not only x[] and y[], but array of t[] which contain time when points was sampled. With time, it could be much more easier...

Yes... I have 4 data inputs. x[] y[] t[] w[], where w[] are the weights.

Maybe you should try parametric fitting? I.e. fit x(t) and y(t) separately, then plot resulting curve. You can use new penalized regression spline fitting functions (described at http://www.alglib.net/interpolation/lea ... #splinefit ).