Thread: nonlinear curve fitting

hi,

i have a rather long non-linear equation with 8 variables that i need to optimize. is there some package/library of ready-made curve fitting c codes that i can use?

thanks!

Gnuplot has some very nice curve fitting tools.

If your objective is to implement a nonlinear model already tested and parameterized, your task is not difficult. However, if you have a set of data, and want to find an equation, I recommend you to give a quick look at the GNU Scientific Library.

Overall, I believe that the most straightforward way to solve your problem is to fit your curve using a standard package (R, GRaphPad Prism), get the equation, and code it in C.

hi, thanks for the replies. i use the curve-fitting solver tool in Excel quite frequently for my other straightforward curve-fitting problems. however, right now, what i need to do is to apply curve-fitting to an image, meaning that i have to iterate this curve-fitting on a pixel by pixel basis (there are about 15million pixels in my image), so Excel is definitely not the way to go.

and yes, i already have a model that i want to optimize, so all i need to do is to vary the model parameters. the model is quite complicated, with exponential, trigonometric, as well as quadratic terms.

what i am hoping to do is to use a readily-available source code for this kind of curve-fitting, and pluck it into my c program. however, i have yet to find any that is easy to understand and implement. there are some links here Levenberg–Marquardt algorithm - Wikipedia, the free encyclopedia but i have yet to try them out.

any other suggestions?

Check -levmar-

Levenberg-Marquardt in C/C++

thanks for the suggestions. i have downloaded the levmar algorithm from the levmar website, and then following the instructions in the FAQ and accompanying readme, i then went on to get lapack. i got lapack from the synaptic package manager (i am running ubuntu 8.04 hardy). as i was not sure which lapack package to get from the list, i installed the 2 that i thought were most likely: liblapack3gf, and, liblapack-dev.

next, i cd-ed into the levmar folder in a terminal, then typed 'make'. this was what i got:

Code:

gcc   -O3 -funroll-loops -Wall    -c -o lm.o lm.c
gcc   -O3 -funroll-loops -Wall    -c -o Axb.o Axb.c
gcc   -O3 -funroll-loops -Wall    -c -o misc.o misc.c
gcc   -O3 -funroll-loops -Wall    -c -o lmlec.o lmlec.c
gcc   -O3 -funroll-loops -Wall    -c -o lmbc.o lmbc.c
gcc   -O3 -funroll-loops -Wall    -c -o lmblec.o lmblec.c
gcc   -O3 -funroll-loops -Wall    -c -o lmbleic.o lmbleic.c
ar crv liblevmar.a lm.o Axb.o misc.o lmlec.o lmbc.o lmblec.o lmbleic.o
a - lm.o
a - Axb.o
a - misc.o
a - lmlec.o
a - lmbc.o
a - lmblec.o
a - lmbleic.o
ranlib liblevmar.a
gcc   -O3 -funroll-loops -Wall    -c -o lmdemo.o lmdemo.c
gcc -L/usr/local/lib  -L. lmdemo.o -o lmdemo -llevmar -llapack -lblas -lf2c  -lm
/usr/bin/ld: cannot find -lf2c
collect2: ld returned 1 exit status
make: *** [lmdemo] Error 1

may i know how to solve this problem?

also, will the algorithm be usable after this, or am i missing some more steps? i would like to try out the expfit.c demo that comes with levmar.

i think i will just use some routines from numerical recipes for my problem. levmar is a bit difficult for me at the moment. thanks for your help.