| Re: [eigen] Status of givens QR decomposition |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
- To: eigen@xxxxxxxxxxxxxxxxxxx, yo.eres@xxxxxxxxx
- Subject: Re: [eigen] Status of givens QR decomposition
- From: Benoit Jacob <jacob.benoit.1@xxxxxxxxx>
- Date: Tue, 26 Jan 2010 06:52:13 -0500
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:received:in-reply-to:references :date:message-id:subject:from:to:content-type :content-transfer-encoding; bh=zf2q8yT5p1DKfJG4nvXlpRf3LHgjG0q7Q6OaR5tTahc=; b=P8Nf3CgTUnpDxgTd9Vq0sGp0ezmQSRFaILtZiz8zeathqWmI+qdA8m90JJnZa3TNp1 ljfMUHNCnsQrNspgaeB3rJ6yQtTgJOuFukA9tI5Y+JWGjuQtlQDFDROKDcHkCv/aUk8Z zkKAMKMYyekk7Rg3Vip1CuEUs8r6hMIHB0Fns=
- Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :content-type:content-transfer-encoding; b=GSocX5kmJz6XZTlwBjyHG6pdxxkB2jVeXSlPKkG6SbUHSmi5AsD4T1oTcx8ax7g65N rtt3BOp7h+p4SUcjzb5uC9q+4UqmeYlbEJDDyFRUI3sCMUSgoMJ5BQ/4cdNM6OsyeJwR 2jUuVdZQzVVsiaYdLYjwN86nOswxCwW0i+LgE=
2010/1/26 Thomas Capricelli <orzel@xxxxxxxxxxxxxxx>:
>
> As a following to:
> http://listengine.tuxfamily.org/lists.tuxfamily.org/eigen/2009/09/msg00009.html
>
> I'd like to say that i have a very important use case for givens-QR with big matrices in the NonlinearOptimization module. This is useful in the variant of the Levenberg Marquardt algorithm for very big matrices, so-called 'storage efficient': the qr decomposition is done with givens rotation, presumably because this is the only decomposition that can be done incrementally : the matrix to decompose is not known as a whole, but is constructed line by line, 'updating' the qr decomposition and deleting the line.
Interesting!!
>
> Andrea, Benoit, what is the status of this fork ? As I have write permission, I've merged with mainline once or twice, but the fork seems abandoned.... was there a use case ? not anymore ?
I haven't touched it in a long time. The original use case was small
matrices, where it is faster than householder. Now you mention large
matrices that are not fully known in advance: interesting.
Finally, about the old debate about whether to store the list of
givens permutations, let me mention that in the different setting of
SVD, I have now decide to try this approach too. It's different as I'm
only storing N Givens rotations per sweep, as opposed to N^2/2 in
GivensQR, but perhaps still worth mentioning.
> btw, Am I the only one needing givens QR decompozition on the list ? anyone ?
It is for sure needed at least by everyone needing QR decomposition of
very small matrices.
If you need to make intrusive changes in Andrea's fork, perhaps it's
easiest for you to fork it, you can then re-merge with Andrea if he
expresses agreement with your changes. I personally don't have time
right now to think about that... :(
Benoit