Re: [rules-users] What's the difference between Drools and other constraints solvers?

Op 22-08-12 03:47, Davide Sottara schreef: For such a small problem you might be better of using brute force (which usually works up to n <= 10). Most solvers don't scale, so any experiment done on a small dataset tells you nothing about how they behave on big datasets :( For example, LocalSolver (the only real competition for Drools Planner), can't handle the roadef B9 instance with 50 000 variables yet with 40GB of RAM as a whole: http://www.localsolver.com/news.html?id=11 Planner can handle it as a whole with only 1GB of RAM and has better results: http://blog.athico.com/2012/06/roadef-2012-first-results-for-dataset-b.html Yes, but Planner doesn't use LP, it uses Metaheuristics. Both have their uses. Metaheuristics and LP can both be applied to planning, scheduling and assignment problems, but in my experience (shown by the competitions), metaheuristics clearly win most battles, especially if the problems scales out. LP is usually better for non-discrete (often non NP-complete) problems, such as figuring out the center of a triangle. LP is also far less powerful on how to define the constraints. For example getting LP to solve a toy problem like TSP requires complex equations (such as comb inequalities) to squeeze the hard constraints into equations. Metaheuristics just repair the chain while doing moves.