[Scipy-tickets] [SciPy] #1588: scipy.optimize.cobyla not consistant in Windows

SciPy Trac scipy-tickets@scipy....
Mon Jan 23 18:35:26 CST 2012


#1588: scipy.optimize.cobyla not consistant in Windows
--------------------------+-------------------------------------------------
 Reporter:  casperskovby  |       Owner:  somebody   
     Type:  defect        |      Status:  needs_info 
 Priority:  normal        |   Milestone:  Unscheduled
Component:  Other         |     Version:  0.10.0     
 Keywords:                |  
--------------------------+-------------------------------------------------
Changes (by pv):

  * status:  new => needs_info


Comment:

 Bugs in f2py sound unlikely, so it's best to rule out other issues first:

 The important question is what are the values of the objective function
 `f(pvec)` -- does this change significantly between runs?

 If it does not change much, then the optimization problem is probably ill-
 defined, and the result is sensitive to rounding error. In that case, this
 is not a bug.

 Now, naively one would expect that the rounding error would be the same
 from one run to another, but I think this depends on the compiler ---
 alignment of data in memory can change, and this can trigger different
 compiler-optimized branches in the code. (See
 http://www.nccs.nasa.gov/images/FloatingPoint_consistency.pdf) If it's
 alignment, then one would expect that the program produces only a few
 different answers. Is this so?

-- 
Ticket URL: <http://projects.scipy.org/scipy/ticket/1588#comment:1>
SciPy <http://www.scipy.org>
SciPy is open-source software for mathematics, science, and engineering.


More information about the Scipy-tickets mailing list