[SciPy-user] Scipy Sparse Library

Dinesh B Vadhia dineshbvadhia@hotmail....
Sat Jan 26 14:34:17 CST 2008


Nathan:

Our version of SciPy is 0.60.win32-py2.5.

We are solving  b = Ax, where b and x are 'float' vectors, A is an MxN sparse matrix (with M != N) with binary (ie. 0 or 1) entries, and is defined as:

        A = scipy.asmatrix(scipy.zeros((M, N), float))            #  'float' because byte int not supported yet

A is populated.  We then use the following SciPy statement to transform into a sparse matrix in csc format:

        A = scipy.sparse.csc_matrix(A)                                 # scipy sparse csc matrix version

Next, calculate b as:

        b = A * x

We don't have drastically varying values.  Cheers!

Dinesh

------------------------------

Message: 6
Date: Sat, 26 Jan 2008 08:07:59 -0600
From: "Nathan Bell" <wnbell@gmail.com>
Subject: Re: [SciPy-user] Scipy Sparse Library
To: "SciPy Users List" <scipy-user@scipy.org>
Message-ID:
<d05265cb0801260607w13c5c011xfd30a1ac03e3dd50@mail.gmail.com>
Content-Type: text/plain; charset=ISO-8859-1

On Jan 25, 2008 9:59 PM, Dinesh B Vadhia <dineshbvadhia@hotmail.com> wrote:
>
> We built a working C++ program using the SparseLib++ library last year.  We
> are now creating a Python version of the program using the Scipy sparse
> library (the csc matrix format) and the results of the program are very
> different to the C++ program.  It's either the Python program or the data.
> We have checked the programs and the data and cannot identify any
> discrepencies.
>
> Is it possible that the Scipy sparse libraries behave differently from the
> SparseLib++ library?  I would have said unlikely but I guess anything is
> possible.

Can you tell us what sparse functionality you are using and which
version of SciPy you have?  Can you produce a short script that
demonstrates the problem?

SciPy sparse arithmetic should produce the same results as
SparseLib++.  The only case when this may not be true is when the
entries of your matrix have drastically varying values (e.g. 1e-10 vs
1e10) and the lack of associativity in floating math produces
different results.

-- 
Nathan Bell wnbell@gmail.com
http://graphics.cs.uiuc.edu/~wnbell/


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://projects.scipy.org/pipermail/scipy-user/attachments/20080126/654dcd50/attachment.html 


More information about the SciPy-user mailing list