|Re: [eigen] inconsistency in fast eigen decomposition|
[ Thread Index |
| More lists.tuxfamily.org/eigen Archives
- To: eigen@xxxxxxxxxxxxxxxxxxx
- Subject: Re: [eigen] inconsistency in fast eigen decomposition
- From: Benoit Jacob <jacob.benoit.1@xxxxxxxxx>
- Date: Thu, 3 Feb 2011 09:00:00 -0500
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:in-reply-to:references:date :message-id:subject:from:to:content-type; bh=ULH6Vnf/K/H371hU1O70H3thUxgS8Iq3PyVcNhM4D6w=; b=s0i1ihxZ2izuYaeiiJdnyhixuyZdT4yWoyY7pZpOJbyajdhGjxC4qitrzZAGzM7QOb t4pT7fOUgxeG84EDIwTR/Z8jpCuJOLWwTQUMoolRqBrIWpyilrxF98N5xbLRcxFNgfm9 HUispDrsw0a/yeFniPa4LpgK7nrTj4IEdqhk8=
- 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; b=VnlWTUHv0w2rB49IPXVFWI7yYEu8fYyhABf6hdSkW1T5D4XG62Ks/1m9JkwjF6uD3D fVTU2VBNQ6DktM16XN8cbW7jj8CH8NGkaPMUJejlMkjJsiSnrq44Gj9pcB4JOWr9pQ8+ j8cNFOBZUXcbWi3oNPrkLIi9j4Rw24y691sQ4=
Oh! I thought that we were talking about a bug in Eigen here. Sorry.
2011/2/3 Gael Guennebaud <gael.guennebaud@xxxxxxxxx>:
> he is not using something which is officially in eigen, but a function
> which is in bench/eig33.cpp that I added for benchmarking purpose and
> possible inclusion in Eigen in the future. The main question is how to
> integrate it (option of SelfAdjointEigenSolver, a new class, both...)
> That can be discussed for 3.1 since there already is a template
> Options parameter in SelfAdjointEigenSolver.
> On Thu, Feb 3, 2011 at 2:24 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
>> Could a unit test be added? Maybe just take his matrix, decompose,
>> reconstruct original matrix, compare. My understanding is that before
>> your fix, eigen would have given a completely wrong reconstructed
>> matrix, so just a fuzzy compare at default test precision would be
>> good enough.
>> 2011/2/3 Gael Guennebaud <gael.guennebaud@xxxxxxxxx>:
>>> yes but was is strange in Radu's example is that the direct method
>>> find a negative eigenvalue while they should all be positive (or equal
>>> to zero). I guess that's because of the multiple trigonometric
>>> functions. Anyway, now I start to compute the eigenvectors from the
>>> biggest eigenvalue and compute the last eigenvectors from the first
>>> two while taking care of some degenerated cases.
>>> On Thu, Feb 3, 2011 at 1:52 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
>>>> Yeah, it's hard to be accurate when there are degenerate eigenvalues.
>>>> Radu's example had well separated eigenvalues.