Re: [R] Non linear modeling

Spencer Graves Sun, 20 Mar 2005 07:57:21 -0800

Another question: What do you know or assume about the distribution of "e"? If (y-x) is always positive, the survival package, especially the survreg function, might help you. For this, I found especially helpful the discussion of this in Venables and Ripley (2002) Modern Applied Statistics with S, 4th ed. (Springer). For a more thorough treatment of the theory (but not of R), see Meeker and Escobar (1998) Statistical Methods for Reliability Data (Wiley).

Alternatively, if you assume a distribution for "e" and assume the different e's are independent, write a function, say, "dev" = (-2) time the sum of the log density function of the e's, then give this to "optim" [e.g., dnorm = density function for a normal distribution; dweibull = density for a Weibull, etc.; help.start() -> "An Introduction to R" covers probability functions and writing your own functions.]

          hope this helps.
          spencer graves

############################
     What do you want to minimize?  Can you write a function to compute
eps given x, y, and a?  Given that, you can then write another function
to compute the objective function you want to minimize.  If "a" is a
scalar, compute the objective function for a range of values of "a" and
plot.  If you want numerical precision, read the help file for "optim",
work the examples until you understand enough to see how to feed your
objective function with a starting value to "optim".

     If you still can't figure it out, please make an attempt, then
read the posting guide "http://www.R-project.org/posting-guide.html";,
and prepare a follow-up question as needed.  (In a discussion on and off
this list earlier this week, several people confirmed that they had
solved many problems following this posting guide.  It may not be as
good as Polya's famous "How to Solve It", but it's pretty good.)

     hope this helps.
     spencer graves

Angelo Secchi wrote:

You are right. eps in my model is not a parameter but the error term.
Also the linearization doesn't solve the problem, since sometimes you
cannot take logs. Any other ideas?
Thanks
On Fri, 18 Mar 2005 11:21:12 -0500
"Liaw, Andy" <[EMAIL PROTECTED]> wrote:
That's treating eps as a parameter in the model. If I read your question right, that's not what you want.
Andy
From: ronggui [mailto:[EMAIL PROTECTED]

then is the nls function can deal the problem as Guillaume STORCHI mentioned in the last post? [X<-nls(y~x+exp(a*x)*eps, data=,start=list(a=,eps=))] or just can solve the problem as:log(y-x) = a*x + e?
On Fri, 18 Mar 2005 08:56:38 -0500
"Liaw, Andy" <[EMAIL PROTECTED]> wrote:
AFAIK most model fitting techniques will only deal with

additive errors, not
multiplicative ones.  You might want to try fitting:
log(y-x) = a*x + e
which is linear.
Andy
From: Angelo Secchi
Hi,
is there a way  in R to fit a non linear model like
y=x+exp(a*x)*eps
where a is the parameter and eps is the error term? Thanks Angelo

______________________________________________ [email protected] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
______________________________________________ [email protected] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html
------------------------------------------------------------------------------ Notice: This e-mail message, together with any attachment...{{dropped}}
______________________________________________ [email protected] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html


______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html

Re: [R] Non linear modeling

Reply via email to