http://www.johndcook.com/blog/2009/01/07/convex-optimization-lectures/ The blog of John D. Cook Sat, 11 Feb 2012 01:10:06 -0500 http://wordpress.org/?v=2.8.4 hourly 1 http://www.johndcook.com/blog/2009/01/07/convex-optimization-lectures/comment-page-1/#comment-11821 gappy Thu, 08 Jan 2009 03:17:47 +0000 http://www.johndcook.com/blog/?p=1225#comment-11821 When it comes to optimization, there are a few rules of thumb I try to follow. First, I try to formulate the problem as a tractable one, i.e., (quasi)convex with a convex feasible region, a linear program, or a discrete linear program (which is non-convex,but for which very effective algorithms exists). Second, I try to find the most parsimonious formulation. This is where experience and skills come in. Last, I never code the algorithm. However simple, existing codes are usually 2-3 times faster than a naive code. There are excellent open-source and commercial codes that do the job, i.e., lpsolve, the <a href="http://www.coin-or.org/" rel="nofollow">COIN-OR</a> family, MOSEK, CPLEX, etc. A lot of them already have python/R/matlab interfaces. When it comes to optimization, there are a few rules of thumb I try to follow. First, I try to formulate the problem as a tractable one, i.e., (quasi)convex with a convex feasible region, a linear program, or a discrete linear program (which is non-convex,but for which very effective algorithms exists). Second, I try to find the most parsimonious formulation. This is where experience and skills come in. Last, I never code the algorithm. However simple, existing codes are usually 2-3 times faster than a naive code. There are excellent open-source and commercial codes that do the job, i.e., lpsolve, the COIN-OR family, MOSEK, CPLEX, etc. A lot of them already have python/R/matlab interfaces.

]]>