[Scipy-tickets] [SciPy] #231: A wrapper for ARPACK

SciPy scipy-tickets@scipy....
Mon Jun 4 15:52:06 CDT 2007


#231: A wrapper for ARPACK
--------------------------+-------------------------------------------------
 Reporter:  nils          |        Owner:  somebody
     Type:  enhancement   |       Status:  new     
 Priority:  normal        |    Milestone:          
Component:  scipy.sparse  |      Version:          
 Severity:  normal        |   Resolution:          
 Keywords:                |  
--------------------------+-------------------------------------------------
Comment (by NeilenMarais):

 OK, had time to have a bit of a look-see. Some comments about the
 interface:

 A) The position of the high level interfaces are a bit wrong. arpack.eigen
 is a generic sparse eigen solver interface, and could conceivably use
 another numerical package to do the job.

 B) Related to A), the routines in arpack.speigs are specific to ARPACK,
 and should be in the main arpack module

 C) Some of the parameters to the high-level interface are a bit ARPACK
 specific. I'm all for a high level interface that is independent of a
 given numeric backend such that a user can pretty much swap them in and
 out without affecting the user's code.

 D) The high level interface should have an easy way of doing spectrum
 shifts, since AFAIK generalized eigen-problems pretty much require them to
 get reasonable efficiency. Not sure how ARPACK specific that would be
 though.

 So, I'd say we should

 1) Name the main module speigs, and have it contain arpack, rather than
 vice versa

 2) Move the high level interface into speigs, and move the arpack specific
 driver functions from speigs to arpack

 3) Think a bit about the high-level interfaces. We should think about the
 best way to talk about solving sparse eigen systems, rather than just use
 "ARPACKisms" for convenience.

 1) and 2) can be done immediately if there is agreement. I could go ahead
 and do it if no-one has any objections. Please poke me if I forget :)

 About 3) I'd love to hear some more input. Perhaps we should discuss it on
 the devel mailinglist?

-- 
Ticket URL: <http://projects.scipy.org/scipy/scipy/ticket/231#comment:19>
SciPy <http://www.scipy.org/>
SciPy is open-source software for mathematics, science, and engineering.


More information about the Scipy-tickets mailing list