I have an experimental set of data ( xdata, ydata) and I want to fit a 5 constant expression to these data and find the unknown constants with following matlab built in function :
[q,resnorm,residual,EXITFLAG,output,LAMBDA]=lsqnonlin('fun',q0)
it converges. but the norm is too big and as of "q", it returns my initial guess as the final fit
. I checked my procedure with some other set of data and a different function, it works perfectly. does this mean that the initial guess is too far? any suggestions?
thanks in adavance
Dear Negar,
Your raw answer would be "Yes, it could". MATLAB nonlinear built-in functions (as for other solvers) heavily rely on initial guess. Most of the time, there is no unique solution to the problem, so, it is not surprising to get different answers by varying the initial guess. However, I suggest to try out other functions in optimization toolbox as well. Specially, trying the GA is an alternative option to check your final output.
Good luck.
It is possible to get a good fit function using heuristic optimization approach. Initially, it is required to guess the function nature using general constants. After that the value of the constants can be found using any heuristic optimization approach.
Alternatively, you could use the output of the first optimization as the next initial guess and repeat the optimization. Repeat this iterative procedure using the output of the nth trial as the input of the (n+1)th trial until the difference between successive outputs is within your desired range.
- Badges
- Science topic