Template talk:Optimization algorithms
Welcome!
Constrained nonlinear optimization
Most of the nonlinear topics include only unconstrained methods. It would be good to have a group for constrained optimization: Sequential unconstrained minimization techniques (SUMT), sequential quadratic programming, augmented Lagrangian methods, Proximal point methods; Successive linear programming (like MS Excel's solver), Gradient projection methods (Lancelot), Michael Saunders's MINOS; Filter methods (Fletcher), etc. Kiefer.Wolfowitz (talk) 23:17, 8 November 2010 (UTC)
Heuristics
Previously, there were 3 lines of heuristics, out of proportion to the treatment of heuristics in optimization textbooks and surveys. There were two other problems, imho.
First, many of the heuristics that were listed have severe problems with notability, just to be listed in Wikipedia, and some have problems with single-purpose accounts (often anonymous IPs) being the main and nearly sole author: such editing is often associated with conflicts of interest, especially self-promotion. I removed all the heuristics with such problems with notability (with even having a WP article).
Second, many of the remaining heuristics were perhaps notable enough to have a Wikidia article, although the concerns with self-promotion and COI remain with a few of them (which seem to share the same editors). However, regardless of their merits as heuristics, many of those heuristics are not described in optimization textbooks and so fail notability to be in the footer for computational optimization. I removed the heuristics that are not discussed by optimization textbooks from this optimization footer.
(It might be useful to create a heuristic footer with them, because they do share a lot of commonalities.)
Finally, I think that my edits are consensus in optimization, because our optimization articles ignore the heuristics that I deleted, which makes it really weird to devote the header to them. I did ask for second opinions at the Wikiprojects in computer science, mathematics, and systems (operations research).
Sincerely, Kiefer.Wolfowitz (talk) 11:03, 30 November 2010 (UTC)
- I don't think branch and bound belongs in the heuristics category as it is/can be complete, unlike the other algorithms in that category. I'd put it in a category on integer linear programming or search algorithms. —Ruud 19:34, 30 November 2010 (UTC)
- Agreed, because the WP article is about branch-and-bound in integer programming. (There are also branch-and-branch methods in global optimization.) Most of this template concerns continuous optimization. It might be worthwhile to consider either moving the name to "continuous optimization methods" or to create at least one line about combinatorial optimization. Kiefer.Wolfowitz (talk) 22:51, 30 November 2010 (UTC)
Combinatorial optimization algorithms
I created a subgrouping for combinatorial optimization algorithms. I was tempted to include mathematical structures important in combinatorial optimization (networks, graphs, matroids, greedoids, etc.) but then this template would not have been focused on algorithms. I did try to include the principal combinatorial algorithms presented in optimization and CS algorithm textbooks. Kiefer.Wolfowitz (talk) 11:43, 2 December 2010 (UTC)
I removed the line on statistical methods. Like the minor heuristics that were removed earlier, the statistical methods are rarely discussed in optimization textbooks (although they are discussed in statistical textbooks, e.g. in system identification).