Parallel rank calculation for sparse matrices -- suggestions?

7 vues (au cours des 30 derniers jours)
Barry
Barry le 30 Août 2022
Commenté : Bruno Luong le 6 Sep 2022
I need to calculate the rank of large (> 1 terabyte of non-zero elements) sparse matrices with MATLAB. Exploring the Parallel Toolbox, but can't seem to find anything that convinces me what is offers will be helpful. If I'm wrong, can someone point me in the right direction? Ideally, I'd just want to take my existing code that has "r=sprank(A)" in it and have that library call run in parallel, perhaps with some additional annotation as needed. Would using a GPUARRAY help here, for example? Doesn't seem like it, but perhaps I'm wrong. Hoping someone here can help me out. Thanks!
Barry Fagin
Professor of Computer Science
barry.fagin@afacademy.af.edu
  9 commentaires
Matt J
Matt J le 5 Sep 2022
A = sparse(tall,skinny)
How tall? How skinny? What currently rules out the approach already discussed below of using rank(full(A.'*A))?
Bruno Luong
Bruno Luong le 6 Sep 2022
"After doing a little digging, I now believe sprank() merely counts the number of non-zero columns. This is not what I require."
It does something more intelligent than that it match the row to column through Dulmange Mendelsohn permutation. In your case might be all the non-zero column can be matched. But that is exactly what structural rank means.

Connectez-vous pour commenter.

Réponse acceptée

Bruno Luong
Bruno Luong le 30 Août 2022
Déplacé(e) : Bruno Luong le 30 Août 2022
Something I don't get : sprank doc tells when you run witth threadpool it runs in parallel.
  5 commentaires
Bruno Luong
Bruno Luong le 31 Août 2022
IMO the doc should clarify this.
But then I still have some doubt. If user runs multiple sprank in parallel, it will take less worker to work on one sprank. So at the end do we really speed up the whole thing?
Edric Ellis
Edric Ellis le 1 Sep 2022
Yes, the "extended capability" description could probably do with some refinement to make it clear exactly what works. (Today, there are a good number of MATLAB functions that cannot be run on a thread pool worker, and this "extended capability" really means simply that the function can run on a thread pool worker).
The performance situation is ... complicated, unfortunately. There is no single simple answer as to whether running N copies of a given function concurrently on workers is faster than running N copies sequentially on the client. It depends on all sorts of details of the implementation of that function - but primarily whether the function is already intrinsically multithreaded by MATLAB itself.

Connectez-vous pour commenter.

Plus de réponses (2)

Matt J
Matt J le 30 Août 2022
Is the matrix square or is it tall and thin? If the later, then it may be an easier computation to compute rank(A.'*A).
  9 commentaires
Barry
Barry le 5 Sep 2022
Modifié(e) : Barry le 5 Sep 2022
I like the A.'*A trick. Or one could ...wait for it ...
... use MATLAB's rref() command, which works on sparse matrices, and then count the nonzero rows. Sigh.
But that in turn begs the question of roundoff errors. Time to move to a new question, which I'll post shortly.
--BF
Matt J
Matt J le 5 Sep 2022
Modifié(e) : Matt J le 5 Sep 2022
I would take caution before relying on rref. It has been often found to lack robustness, see e.g.,

Connectez-vous pour commenter.


Bruno Luong
Bruno Luong le 5 Sep 2022
For thin and tall sparse matrix A of size (m x n), m>>n and n in the order of 1000s, it might be possible to compute the rank using q-less qr, which is better than rank(A'*A) which has the drawback of square the condition number.
A=sprand(100000,10,0.1)*sprand(10,100,0.1);
R=qr(A,0);
rank(full(R))
ans = 10

Catégories

En savoir plus sur Linear Algebra dans Help Center et File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by