Hey. I was wondering if anybody here has a suggestion for a good curve fitting algorithm. I would prefer something that is computationally fast, or that does not need to be completely recomputed every new point I wish to fit. Thanks
FANN is very efficient. You didn't make clear if you were referring to a software library or just algorithms themselves. You can find the algorithms used by fann in it's source code if desired.
Either an algorithm or library would do. Iv never heard of NNs being used for curve fitting. I was thinking of something along the lines of regression analysis, cubic spline fitting, or polynomial fitting.
Matlab has a whole toolbox for curve-fitting. Can't say I have much experience with that toolbox, however.
Curve fitting is a classification of various types such as you've mentioned. NNs handle it very well (too much so for data classification, etc. if you're not careful ). A single intermediary node will produce a linear regression, multiple node models will regress non-linearly. Here's a nice summary definition of curve fittng. http://www.answers.com/main/ntquery?s=curve+fitting&gwp=13 And one for NNs. http://www.answers.com/topic/artificial-neural-network
to clarify a bit. I am coding in c++ if anyone knows good public / open source libraries. Ill take a look at FANN, even though I dont know off hand how to apply it to curve fitting.
Well it's actually one in the same. But if you you'd like something more direct, take a look at R, Octave, Scilab, etc. which are real good open source statistics software application suites. But they use an interpreter. Also, here's a nice on-line reference for stat related stuff. Look at regression modeling. StatSoft