nls, restrict parameter values

Thu, May 8, 2003 8:38 AM #

Hi,
I posted a question (bellow) a few weeks ago and had a reply (thanks 
Christian) that partly solves the problem, but I still would like to be able 
to restrict some of the independent variables in a nls model to be always 
 >0, (is there a way to do it)??
Thanks,
Angel

From: "Christian Ritz" <ritz at dina.kvl.dk>
To: "Angel -" <angel_lul at hotmail.com>
CC: <r-help at stat.math.ethz.ch>
Subject: Re: [R] nls: Missing value or an Infinity produced when evaluating 
the model
Date: Wed, 23 Apr 2003 15:22:37 +0200

Hi Angel,

I tried reparametrise your model, setting:

BirthMass^0.25=u
MaxMass^0.25=v

and giving the following formula in R:

GrowthModel<-nls(BodyMass~(((1-(1-u/v)*exp(-a*Time/(4*v)))^4)*v^4),data=grow
th,start=c(u=4,v=5,a=1.5),trace=TRUE)

And this works for me, but the u estimate is negative (not significantly
different from 0, though).

Christian

----- Original Message -----
From: "Angel -" <angel_lul at hotmail.com>
To: <r-help at stat.math.ethz.ch>
Sent: Wednesday, April 23, 2003 2:29 PM
Subject: [R] nls: Missing value or an Infinity produced when evaluating the
model

Hi,
I am trying to fit a sigmoid curve to some data with nls but I am

getting

into some trouble.
Seems that the optimization method is getting down to some parameter
estimates that make the equation unsolvable. This is an example:

growth<-data.frame(Time=c(5,7,9,11,13,15,17,19,21,23,25,27),BodyMass=c

(45,85,125,210,300,485,570,700,830,940,1030,1120))

GrowthModel<-nls(BodyMass~(((1-(1-((BirthMass/MaxMass)^0.25))*exp(-a*Time/

(4*MaxMass^0.25)))^4)*MaxMass),data=growth,start=c
(BirthMass=3,MaxMass=2500,a=1.5),trace=TRUE)

56043.86 :     3.0 2500.0    1.5 >Error in numericDeriv(form[[3]],

names(ind), env) : >        Missing value or an Infinity produced when 
evaluating the model

Is there anyway I can restrict the parameter values used so it doesn't

get to this no return point.

Any other alternatives are also very welcome!
Thanks in advance,
Angel

_________________________________________________________________
Use MSN Messenger to send music and pics to your friends 
http://www.msn.co.uk/messenger

Spencer Graves

Thu, May 8, 2003 9:08 AM #

If I have a parameter I want to be positive, I replace it by its 
logaritm.  That sends 0 to (-Inf).  Often when I do this, I also get a 
log(likelihood) that is more closely parabolic, which reduces parameter 
effects and makes the standard normal approximations more accurate.

hth.  spencer graves

Angel - wrote:

Hi,
I posted a question (bellow) a few weeks ago and had a reply (thanks 
Christian) that partly solves the problem, but I still would like to be 
able to restrict some of the independent variables in a nls model to be 
always >0, (is there a way to do it)??
Thanks,
Angel

From: "Christian Ritz" <ritz at dina.kvl.dk>
To: "Angel -" <angel_lul at hotmail.com>
CC: <r-help at stat.math.ethz.ch>
Subject: Re: [R] nls: Missing value or an Infinity produced when 
evaluating the model
Date: Wed, 23 Apr 2003 15:22:37 +0200

Hi Angel,

I tried reparametrise your model, setting:

BirthMass^0.25=u
MaxMass^0.25=v

and giving the following formula in R:

GrowthModel<-nls(BodyMass~(((1-(1-u/v)*exp(-a*Time/(4*v)))^4)*v^4),data=grow 

th,start=c(u=4,v=5,a=1.5),trace=TRUE)

And this works for me, but the u estimate is negative (not significantly
different from 0, though).

Christian

----- Original Message -----
From: "Angel -" <angel_lul at hotmail.com>
To: <r-help at stat.math.ethz.ch>
Sent: Wednesday, April 23, 2003 2:29 PM
Subject: [R] nls: Missing value or an Infinity produced when 
evaluating the
model

Hi,
I am trying to fit a sigmoid curve to some data with nls but I am

getting

into some trouble.
Seems that the optimization method is getting down to some parameter
estimates that make the equation unsolvable. This is an example:

growth<-data.frame(Time=c(5,7,9,11,13,15,17,19,21,23,25,27),BodyMass=c

(45,85,125,210,300,485,570,700,830,940,1030,1120))

GrowthModel<-nls(BodyMass~(((1-(1-((BirthMass/MaxMass)^0.25))*exp(-a*Time/

(4*MaxMass^0.25)))^4)*MaxMass),data=growth,start=c
(BirthMass=3,MaxMass=2500,a=1.5),trace=TRUE)

56043.86 :     3.0 2500.0    1.5 >Error in numericDeriv(form[[3]],

names(ind), env) : >        Missing value or an Infinity produced when 
evaluating the model

Is there anyway I can restrict the parameter values used so it

doesn't get to this no return point.

Any other alternatives are also very welcome!
Thanks in advance,
Angel

_________________________________________________________________
Use MSN Messenger to send music and pics to your friends 
http://www.msn.co.uk/messenger

______________________________________________
R-help at stat.math.ethz.ch mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-help