Institut für Astronomie und AstrophysikAbteilung AstronomieSand 1, D-72076 Tübingen, Germany |
error
compute errors for fit using the chi^2 distribution.
cafe
error [,parameter][,iter][,delchi][,/verbose][,/selected,/iterstop]
parameter - (optional) The parameter to compute the error for. Default are all parameters. Allowed are parameter indices, ranges, names with wildcards "*". iter - (optional) Number of iterations to perform before query to prolong. delchi - (optional) Delta chi^2 to be used for the determination of the uncertainty. Default: 1, corresponding to 1 sigma errors. Use 2.71 for 90% uncertainty (in general, delchi=chisqr_dvf(1-prob,1) where prob is the probability that the n dimensional parameter space spanned by the uncertainties contains the real value). verbose - Show fit process progress more or less elaborated. This may be a number according: 0 - quiet 1 - basic information (default) 2 - additional boundary information 3 - additional progress information 4 - fit processing information selected - Compute error for selected data points only. nodata - Take also models into account which are in groups containing no data. May be usefull when building complex models refering one to other. iterstop - Stop when key 'Q' pressed. Default true.
All inputs/options (except parameter) may be set with the "set" command. The command prefix is "error".
This command tries to estimate the error by stepping through the Chi^2 distribution and getting the confidence range for a certain delta chi (s.a. delchi). This approach is more error prone than using the hessian values from the fit command. Sometimes this error computation fails. In this case it is recomended to use the steppar command and analyse the resulting Chi^2 distribution manually. When performing the error process the given parameter is stepped while leaving the other free for fit. For these parameters apply the following rules: - if a parameter is frozen (fixed flag = 1) the parameter will not be touched during the error computation. - if a parameter is tied to another parameter this parameter will copy the latter parameter value (in case of expressions the value will be mapped with this expression). - groups which contain a model but no valid data points are ignored. In case of the selected flag (s.a.) only selected data points are taken into account, i.e. if no data points are selected the group is ignored. - groups which contain valid data points but no model are ignored.
Changes parameter error values. Because the confidence range may be asymetric both lower and upper limit are stored. For reference the larger deviation is given.
The error determination may fail if the previous fit contained bad hessian error estimations. It is recomended to run the fit process before. Long lasting fit processes may be interrupted with "Ctrl-G". (Not possible in idlwave).
> model, "parabel", 1 > fit > error -> compute errors for parameter
$Id: cafe_error.pro,v 1.24 2004/07/01 13:07:56 goehler Exp $
[Home Page] [Software, Documentation] [IDL Documentation] [Quick Reference] [Feedback]