[Scipy-tickets] [SciPy] #808: fit method in stats.distribution needs better starting values

SciPy scipy-tickets@scipy....
Mon Dec 1 11:58:03 CST 2008


#808: fit method in stats.distribution needs better starting values
-------------------------+--------------------------------------------------
 Reporter:  josefpktd    |       Owner:  somebody
     Type:  enhancement  |      Status:  new     
 Priority:  normal       |   Milestone:  0.8     
Component:  scipy.stats  |     Version:          
 Severity:  normal       |    Keywords:          
-------------------------+--------------------------------------------------
 For some distributions the default starting values for the maximum
 likelihood estimation in fit don't make any sense, e.g.

 `truncnorm`: starting values for lower and upper bound is equal to one. As
 a consequence fit doesn't move away from the starting value.

 A new method, that is called by fit and can be overwritten by the
 individual distributions, and that determines the starting values from the
 data would allow a better "automatic" estimation.

 Another example:
 From what I remember during testing: `genextreme`, in some cases, only
 converges to correct parameters if the starting value is smaller than the
 true value.

-- 
Ticket URL: <http://scipy.org/scipy/scipy/ticket/808>
SciPy <http://www.scipy.org/>
SciPy is open-source software for mathematics, science, and engineering.


More information about the Scipy-tickets mailing list