[Neurodebian-devel] [psignifit-devel] [ftpmaster at debian.org: psignifit_3.0~beta.20120503.1-1_amd64.changes REJECTED]

[Hänel Nikolaus Valentin]

* Yury V. Zaytsev <yury at shurup.com> [2012-05-23]:
> On Wed, 2012-05-23 at 15:04 +0200, Hänel Nikolaus Valentin wrote:
> > 
> > Well... I guess so.. I still think pypsignifit is not a good idea,
> > since the source package may at somepoint also provide mpsignifit,
> > rpsignifit and psignifit-cli. 
> 
> Right, so how much work is it to make it provide all functionality of
> psignifit and obsolete the latter? Is it even feasible?

Well, the problem is, he optimizers are slightly different. In practice,
psignifit-ng (lets just call it that for now) provides all the features
of psignifit and more. However certain individuals are interested in
comparing the two optimizers for speed and correctness.

You know how much numerical optimization can be a voodoo in practice?!
Determining start values, transformation of goal function to avoid
expensive operations, penalty terms, numerical accuracy, restarting
inside minimum, possible local minima and and and..

It will be very, very, very hard be to replicate the optimizer of the
old psignifit _EXACTLY_...

> Otherwise you can go for psignifit-ng ;-)

Yeah, I quite like this. Or psignifit3

V-