STScI Logo

controlpars stsdas.analysis.fitting


NAME · USAGE · DESCRIPTION · PARAMETERS · SEE_ALSO

NAME

controlpars -- Edit parameters related to the non-linear fitting algorithm.

USAGE

controlpars

DESCRIPTION

The basic parameter is method, which allows selection of the chi-square minimization method to be used. Currently the package supports two methods: amoeba (downhill simplex) and Levenberg-Marquardt.

The amoeba minimization algorithm looks for a minimum of the chi-square between function and data using a systematic, iterative, albeit time-consuming, direct search in coefficient space. It does not use derivatives of the function being fitted, but just the chi-square as a function of the coefficients. The Levenberg-Marquardt method uses a more subtle technique that relies on the function derivatives to optimize its search steps. Because of that it is in principle faster than the amoeba method, but the derivative computation can introduce instabilities that may impair convergency.

Because both methods are iterative, it is advisable to restart the fit once they find a minimum, restarting from the last point found in coefficient space. This might help the methods to escape from secondary minima.

Two parameters are provided to control the iteration sequence. maxit is the maximum number of iterations before the algorithm attempts a restart. restart is the maximum number of restarts. If execution time is at a premium, it is advisable to allow at least a couple of restarts, at the expense of a smaller number of iterations between restarts.

The three parameters alpha, beta' and gamma apply only to the amoeba method, and control the relative amount of simplex expansion/contraction when searching coefficient space. Care must be exercised when changing these parameters. They should be modified only when very erratic behavior is detected in the fitting process, and even so only by small (~20%) amounts.

PARAMETERS

(method = "amoeba") [char]
Allowed values: amoeba | Levenberg-Marquardt. Selects the minimization method.
(maxit = 500) [integer, min=10]
Maximum number of iterations in between restarts.
(restart = 3) [integer, min=1]
Maximum number of restarts.
(alpha = 1.0, beta = 0.5, gamma = 2.0) [real]
Parameters for controlling simplex expansion/contraction.

SEE ALSO

nfit1d, ngaussfit, n2gaussfit


Package Help · Search Form · STSDAS