[R] optimizing with non-linear constraints

[Rainer M Krug]

Envoyé de mon iPhone

> Le 1 oct. 2015 à 15:17, Ravi Varadhan <ravi.varadhan at jhu.edu> a écrit :
> 
> I would recommend that you use auglag() rather than constrOptim.nl() in the package "alabama."  It is a better algorithm, and it does not require feasible starting values.

Thanks - that was one question I wanted to ask later. 

I will do so,

Rainer

> Best,
> Ravi  
> 
> -----Original Message-----
> From: Rainer M Krug [mailto:Rainer at krugs.de] 
> Sent: Thursday, October 01, 2015 3:37 AM
> To: Ravi Varadhan <ravi.varadhan at jhu.edu>
> Cc: 'r-help at r-project.org' <r-help at r-project.org>
> Subject: Re: optimizing with non-linear constraints
> 
> Ravi Varadhan <ravi.varadhan at jhu.edu> writes:
> 
>> Hi Rainer,
>> It is very simple to specify the constraints (linear or nonlinear) in 
>> "alabama" .  They are specified in a function called `hin', where the 
>> constraints are written such that they are positive.
> 
> OK - I somehow missed the part that, when the values x are valid, i.e. in the range as defined by the conditions, the result of hin(x) that they are all positive.
> 
>> Your two nonlinear constraints would be written as follows:
>> 
>> hin <- function(x, LAI) {
>> h <- rep(NA, 2)
>> h[1] <- LAI^x[2] / x[3] + x[1]
>> h[2] <- 1 - x[1] - LAI^x[2] / x[3]
>> h
>> }
> 
> Makes perfect sense.
> 
>> 
>> Please take a look at the help page.  If it is still not clear, you can contact me offline.
> 
> Yup - I did. But I somehow missed the fact stated above.
> 
> I am using constrOptim() and constrOptim.nl() for a paper and am compiling a separate document which explains how to get the constraints for the two functions step by step - I will make it available as a blog post and a pdf.
> 
> I might have further questions concerning the different fitting functions and which ones are the most appropriate in my case.
> 
> Thanks a lot,
> 
> Rainer
> 
> 
>> Best,
>> Ravi
>> 
>> Ravi Varadhan, Ph.D. (Biostatistics), Ph.D. (Environmental Engg) 
>> Associate Professor,  Department of Oncology Division of Biostatistics 
>> & Bionformatics Sidney Kimmel Comprehensive Cancer Center Johns 
>> Hopkins University
>> 550 N. Broadway, Suite 1111-E
>> Baltimore, MD 21205
>> 410-502-2619
>> 
>> 
>>    [[alternative HTML version deleted]]
>> 
> 
> --
> Rainer M. Krug
> email: Rainer<at>krugs<dot>de
> PGP: 0x0F52F982