Have you considered fitting on the log scale? It's not clear to
me what problem you are trying to solve, but the following will estimate
parameters in a model that looks to me like what you are describing:
lm(z ~ (y1+y2+...+yn)^n, ...)
hope this helps.
spencer
Zhen Pang wrote:
> Dear all,
>
> I wonder if there are some function or package in R which can do the
> iterative proportional fitting.
>
> In the exponential model
> f(y1,...yn)=exp(a'yi+b'(yi*yj)+.....+c'(y1*...*yn)+constance),
>
> instead of the canonical parameters, I use maginal probability instead
> of a and log odds ratio instead of b. and for the order higher than 3,
> I use the canonical way or even assume them to be 0.
>
> It is good if there is direct function or package in R to do the job,
> which will save a lot of time of me. Thank you!
>
> Rgs,
>
> Zhen