[Scipy-tickets] [SciPy] #1553: Improve distribution fitting

SciPy Trac scipy-tickets@scipy....
Thu Nov 3 11:01:12 CDT 2011


#1553: Improve distribution fitting
-------------------------+--------------------------------------------------
 Reporter:  pbrod        |       Owner:  somebody   
     Type:  enhancement  |      Status:  new        
 Priority:  normal       |   Milestone:  Unscheduled
Component:  scipy.stats  |     Version:  devel      
 Keywords:               |  
-------------------------+--------------------------------------------------

Comment(by josefpktd):

 good idea, I think this will only be relevant if the initial conditions
 have points outside of the support, but we still have several of those.
 Until now I only looked at how we could improve the starting values, but
 this change will also find them (at least in some cases).

 I guess I roughly understand the choice of 100*log(realmax).

 One caveat, I think there are still problems with having the correct
 bounds a and b in some distributions. (independent of the proposed change
 here)

 For example, genpareto only defines b in argcheck, tukeylambda doesn't set
 a or b.
 In these cases the out of support inf is created by the logpdf method (if
 everything is correct). (I didn't check the details carefully.)

-- 
Ticket URL: <http://projects.scipy.org/scipy/ticket/1553#comment:1>
SciPy <http://www.scipy.org>
SciPy is open-source software for mathematics, science, and engineering.


More information about the Scipy-tickets mailing list