Changeset b7ce5ad in sasview


Ignore:
Timestamp:
Mar 28, 2019 12:11:47 PM (4 months ago)
Author:
smk78
Branches:
master, magnetic_scatt
Children:
a48831a8
Parents:
e40390ee
Message:

Added section on Fitting Integer Parameters. Closes #882

File:
1 edited

Legend:

Unmodified
Added
Removed
  • src/sas/sasgui/perspectives/fitting/media/fitting_help.rst

    r9258c43c rb7ce5ad  
    338338These optimisers form the *Bumps* package written by P Kienzle. For more information 
    339339on each optimiser, see the :ref:`Fitting_Documentation`. 
     340 
     341.. ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ 
     342 
     343Fitting Integer Parameters 
     344-------------------------- 
     345 
     346Most of the parameters in SasView models will naturally take floating point (decimal)  
     347values, but there are some parameters which can only have integer values. Examples  
     348include, but are not limited to, the number of shells in a multilayer vesicle, the  
     349number of beads in a pearl necklace, the number of arms of a star polymer, and so on. 
     350Wherever possible/recognised, the integer nature of a parameter is specified in the  
     351respective model documentation and/or parameter table, so read the documentation  
     352carefully! 
     353 
     354Integer parameters must be fitted with care. 
     355 
     356Start with your best possible guess for the value of the parameter. And using  
     357*a priori* knowledge, fix as many of the other parameters as possible. 
     358  
     359The SasView optimizers treat integer parameters internally as floating point  
     360numbers, but the values presented to the user are truncated or rounded, as  
     361appropriate. 
     362 
     363In most instances integer parameters will probably be greater than zero. A good  
     364policy in such cases is to use a constraint to enforce this. 
     365 
     366Because an integer parameter should, by definition, only move in integer steps,  
     367problems may be encountered if the optimizer step size is too small. Similarly,  
     368be **very careful** about applying polydispersity to integer parameters. 
     369 
     370The Levenberg-Marquardt and Quasi-Newton BFGS (and other derivative-based)  
     371optimizers are probably best avoided for fitting models with integer parameters. 
    340372 
    341373.. ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ 
Note: See TracChangeset for help on using the changeset viewer.