Boost logo

Ublas :

Subject: Re: [ublas] Matrix decompositions.
From: oswin krause (oswin.krause_at_[hidden])
Date: 2013-05-15 01:53:14


in the order I stumbled over the things:

line 44-54: you don't need a copy, instead you should use a combination
of row/subrange.
line 58-60: you should take a look at inner_prod
line 63: 0.00025 is too big.
line 66: You should never create H explicitly.
line 67: because you formed H, this step is O(n^3) which makes the whole
algorithm O(n^4). This can be done in O(n^2)
line 73-79: same applies here.

On 14.05.2013 22:12, Salman Javaid wrote:
> Hello uBLAS Contributors:
> I have applied to GSoC 2013 and
> pitched implementation of SVD factorization for uBLAS. In order to
> better prepare myself and to get my hands dirty at uBLAS, I ended up
> implementing QR Factorization employing Householder Reflections using
> uBLAS. This is only the first draft and will be needing significant
> improvement, e.g., computation of QR decomposition of 100 * 100 matrix
> takes around 30 seconds. But I guess just to get familiar with code
> base, it was a good exercise. Over the next week or two I will be
> trying to optimize the code.
> I will be absolutely grateful if
> contributors can have a quick glance at the code, and point me to any
> improvements they can suggest. Particularly in speeding up matrix
> multiplication.
> I used Visual Studio 2010 to compile the code. I will try to get the
> code running on my Ubuntu machine in a couple of days hopefully.
> Here the header file:
> The main file:
> Best Regards,
> Salman Javaid
> _______________________________________________
> ublas mailing list
> ublas_at_[hidden]
> Sent to: Oswin.Krause_at_[hidden]