R help - Jan 2009 - constrainOptim

Dear R helpers
 
I have a question regarding the constrainOptim. 
I'm coding the nested logit and would like to set a bound of rho to (0,1] as
an extreme value distribution where rho = exp(lambda)/1+exp(lambda)
I wonder if I can do that directly in optim (say rho > 0 & <= 1) or
need to use constrainOptim
I read the help but still don't know how to set ui and ci
 
Thanks,
June

_________________________________________________________________


	[[alternative HTML version deleted]]

For simple box constraints, i.e. lower and upper limits directly on the
parameters themselves, you don't need ConstrOptim.  You can get the job done
with the "L-BFGS-B" algorithm in optim() or using nlminb() or using
the spg() function in the BB package.  In this case the feasible region is a
hyper-rectangle (could be infinite in some dimensions).

ConstrOptim() is useful when you have more general linear inequality
constraints, i.e. constraints on linear combinations of parameters. In this case
the feasible region is a convex polytope.

Best,
Ravi.

____________________________________________________________________

Ravi Varadhan, Ph.D.
Assistant Professor,
Division of Geriatric Medicine and Gerontology
School of Medicine
Johns Hopkins University

Ph. (410) 502-2619
email: rvaradhan at jhmi.edu


----- Original Message -----
From: June Wong <neptune545 at hotmail.com>
Date: Wednesday, January 28, 2009 11:57 am
Subject: [R] constrainOptim
To: r-help at r-project.org

>  Dear R helpers
>   
>  I have a question regarding the constrainOptim. 
>  I'm coding the nested logit and would like to set a bound of rho to 
> (0,1] as an extreme value distribution where rho =
exp(lambda)/1+exp(lambda)
>  I wonder if I can do that directly in optim (say rho > 0 & <= 1)
or
> need to use constrainOptim
>  I read the help but still don't know how to set ui and ci
>   
>  Thanks,
>  June
>  
>  _________________________________________________________________
>  
>  
>  	[[alternative HTML version deleted]]
>  
>  ______________________________________________
>  R-help at r-project.org mailing list
>  
>  PLEASE do read the posting guide 
>  and provide commented, minimal, self-contained, reproducible code.

June Wong <neptune545 <at> hotmail.com> writes:
> 
> 
> Dear R helpers
> 
> I have a question regarding the constrainOptim. 
> I'm coding the nested logit and would like to set a bound of rho to
(0,1] as
an extreme value distribution
> where rho = exp(lambda)/1+exp(lambda)
> I wonder if I can do that directly in optim (say rho > 0 & <= 1)
or need to
use constrainOptim
> I read the help but still don't know how to set ui and ci
> 
> Thanks,
> June
> 
  optim() can do box constraints (i.e., independent inequality
constraints on parameters): use method="L-BFGS-B" and
the lower and upper arguments to set the bounds for
each parameter (to -Inf and Inf if there are no bounds).
If you want to set bounds on rho you have to use rho as
the parameter in your model -- this is tricky if you
can't solve for rho, but in your case lambda=log(rho/(1-rho))

  Ben Bolker