[R] Optim simplex start size problem

Scott Brown scottb at uci.edu
Thu Jan 13 20:14:46 CET 2005


Hi R Users,

I've struck a problem with 'optim', which has previously worked very 
well for me.  I cannot find a way to set the size of starting simplex 
when using the Nelder Mead method.  The "parscale" argument will control 
the relative sizes of sides, but not the global magnitude.

A short example:

f<-function(x){print(x);sum(sin(x))}  # A minimum at (0,0).
optim(par=c(1,1),fn=f,method="Nel",control=list(maxit=10,parscale=c(1,1))

That makes the initial simplex look like this:
1,   1
1.1, 1
1,   1.1

Now say I want to use a larger intial simplex, mabye:
1,   1
2,   1
1,   2

The documentation suggests that parscale=c(10,10) should do it.  But if 
I do that, I get the same simplex as before.

Only the *relative* sizes of parscale seem to matter (I can make one 
side longer than the others).

Does anyone have any advice for me?

Scott Brown.




More information about the R-help mailing list