Re: RFC: quadprog/lsqlin with __qp__ (Re: GSoC 2015, optim)

[Olaf Till]

On Thu, Aug 27, 2015 at 12:44:18PM +0100, Asma Afzal wrote:
> > I've only roughly looked at the part where you handle the lambda
> > returned by __qp__, but there seems to be an issue with the equality
> > constraints resulting from too close upper and lower bounds. You move
> > this lambda away from equality constraints to the upper bounds lambda,
> > which seems correct, but should it not also be copied to the lower
> > bounds lambda at the same index?
> 
> I checked this with Matlab. It puts the multipliers for too close
> bound constraints in the corresponding indices of lambda.lower. I have
> done the same now.
> Should I put it in lambda.upper as well?

This would seem logical to me, but if Matlab doesn't do it ... I'm not
sure, but since I can't see any reason but Matlab compatibility to
return lambda at all, I'd say do it as Matlab does it, i.e. as you do
it now.

I'll look at your new code soon.

Olaf

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

signature.asc
Description: Digital signature