forum.alglib.net
http://forum.alglib.net/

RBF algorithm clarifications
http://forum.alglib.net/viewtopic.php?f=2&t=1788
Page 1 of 1

Author:  tiggey [ Mon May 05, 2014 11:19 am ]
Post subject:  RBF algorithm clarifications

Hi all. First off sorry for my poor english. Second... Thank you very much for this fantastic project. Alglib seems to be powerful and efficient and it must be a lot of work behind it, sure.

I'm an italian student and I'm using Alglib in order to interpolate some metereological information. I.e. I have some scattered wheater points, that could be scalar (such as temperature) or vector. These point are not gridded and the distance between them is random. I'd like to obtain a prediction of these information on arbitrary points in the 3d space.

I was looking for something interesting and I've tried to use alglib in order to solve this problem, but:

- my idea was to use an inverse distance weighted interpolation, but i see that is considered deprecated and replaced by RBF algorithms.

- I've tried to use RBF algorithms, but I can't obtain the expected result. I've tried QNM and Multilayer algorithms and I was stuck in the examples. I don't realize why, for example, when i try to use the example of multilayer, when I try to evaluate the function in points placed so far from the scattered-input-data-points (i.e. 100,100) the prediction on these points is so far from scattered values. Evaluating the function in 100,100, for example, I have a result of -9.60. But the samples are defined by

real_2d_array xy0 = "[[-2,0,1],[-1,0,0],[0,0,1],[+1,0,-1],[+2,0,1]]";

is it possible that prediction could be so far? This result is independent from the number of iterations that i use.

Is it a normal behaviour? Have I to use some different algorithm in order to solve my problem? Splines? Or deprecated version of inverse weighted prediction?


Thank you very much for your patience.

S.

Page 1 of 1 All times are UTC
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/