Re: cauchy.m in optim package?

[Olaf Till]

On Thu, Jul 24, 2014 at 06:25:06PM +0200, Juan Pablo Carbajal wrote:
> On Thu, Jul 24, 2014 at 9:30 AM, Olaf Till <address@hidden> wrote:
> > Though I don't completely understand its algorithm, cauchy.m does not
> > seem to belong into the optim package for me. AFAICT it computes
> > derivatives (even if also those of higher order) only with respect to
> > a single parameter. And even the returned first derivatives vary
> > vastly with different values for the arguments `N' and `r', as simple
> > tests with sin(x) for x=0 show.
> >
> > Both issues seem to make it fairly useless for optimization. If nobody
> > corrects me, I'd suggest to remove it from the optim package.
> >
> > Olaf
> >
> > --
> > public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net
> 
> Hi Olaf,
> 
> Her eis the reasoning why we add it there. Basically other functions
> to calculate numerical derivatives are already in optim, so if you are
> not against those, I wouldn't know why you are against this one there.

Hi Juan Pablo,

but havn't I detailed above why I'm "against this one" (in optim) and
not against the others? The others return gradients, not only single
derivatives, and they do it rather exactly without chosing suitable
`N' and `r'.

> http://octave.1599824.n4.nabble.com/Taylor-expansion-using-the-fft-td3885008.html#a3885077

I see nothing in the above thread which addresses my two concerns.

Olaf

-- 
public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net

signature.asc
Description: Digital signature