[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Help-gsl] ways to change sensitivity of the Lev enberg–Marquardt so
Re: [Help-gsl] ways to change sensitivity of the Lev enberg–Marquardt solver to the initial guess
Mon, 20 Dec 2010 16:17:25 +0100
> For no-noise data and perfect start guesses, both implementations fit
> perfectly (you should hope so). When changing the guess just slightly
> the lmsder fitter fails, whereas the simplex routine is robust for
> much larger differences between the initial guess and the solution.
> What can I do to improve the performance of the lmsder - is there any
> tuning I can do to my function or the solver? (Improving the guess to
> what appears to be required from lmsder will not be possible).
as far as I know the lmsder algorithm inherently needs "good" initial
guess of the parameters. If you needs a more robust method you can
perform a preliminary search with the simplex minimizer and, when you
have a reasonable starting point, you can feed it to the lmsder
algorithm. The advantage of the lmsder is that it will converge much
more rapidly toward the optimal solutions.
I hope that helps.