search

techoe / ceres-solver

Automatically exported from code.google.com/p/ceres-solver
Other
0 stars 0 forks source link

User orderings in SPARSE_NORMAL_CHOLESKY #88

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago 
Currently, the SPARSE_NORMAL_CHOLESKY solver uses AMD to determine he best fill 
reducing ordering. Finding the optimal fill reducing
ordering is NP-Hard, so AMD uses a bunch of heuristics. For a lot of problems 
this is fine. But for some problems, the user knows the problem structure 
before hand and can help AMD find a better ordering.

Work: We already have the infrastructure for the user to specify the ordering 
using the ParameterBlockOrdering object. The work centers around using this 
object to call CAMD instead of AMD. 4-5 hours of work.

Original issue reported on code.google.com by sameerag...@google.com on 3 Mar 2013 at 9:40

GoogleCodeExporter commented 9 years ago 
There seems to be a bug in the way SuiteSparse is compiled. If metis not 
present, the cholmod_camd is not available. Pinged Prof. Davis about this.

Original comment by sameerag...@google.com on 4 Mar 2013 at 2:30

GoogleCodeExporter commented 9 years ago 
Prof. Davis indicates that he will push out a new version of SuiteSparse with 
this issue fixed.

Original comment by sameerag...@google.com on 12 Mar 2013 at 12:37

GoogleCodeExporter commented 9 years ago 
Tried this on my local machine, and it doesn't really help much. The basic AMD 
does a pretty good job without user hints. But that may be a function of the 
problems I am playing with rather than the approach itself.

Still adding this right now without widespread support in suitesparse will lead 
to compile time problems, and I would like to avoid that complexity. So making 
it won't fix for now.

Original comment by sameerag...@google.com on 16 Mar 2013 at 4:38