Thanks you've helped lots!

Genetic Algorithms is one of the best hill climbing method.

QuoteOriginally posted by: cybernicusGenetic Algorithms is one of the best hill climbing method.Except for almost all the others. I've never encountered a problem for which GA were actually the best solution. In any given case there always seemed to be something else that worked better once the problem had been properly understood.

QuoteOriginally posted by: zarnywhoopQuoteOriginally posted by: cybernicusGenetic Algorithms is one of the best hill climbing method.Except for almost all the others. I've never encountered a problem for which GA were actually the best solution. In any given case there always seemed to be something else that worked better once the problem had been properly understood.That is the crunch of it really, understanding the problem. If the "better" method, gives the same answer, but quicker, it could be argued that it is better to get a GA to do the work of "understanding" the problem than doing that by hand. This could particularly be true if you are going to do a lot of similar problems, so you understand the overall family, but not the intricate details of each specific case.Of course problems such as local minima etc HAVE to be sorted out and understood because you get the wrong answer otherwise.

QuoteOriginally posted by: MikeCroweAm just wondering what experiances people have with different minimisation algorithms... efficiency etc.I'm really looking at high dimension (1000+) multimodal functions (think backpropagation nets) and I'm struggling to get good convergence.Any suggestions/citations?I have been playing with PSO - particle swarm optimization and it seems to have good convergence on some test problems..rosenbrock's function and De-Jong's test function. I think random search combined with determinism is the way to go in such a high dimensional problem? I will post my r-code of PSO here soon.

I second genetic algorithms (GAs). I dont want platform you are working on but I think MSDN has come out with a GAs toolbox.QuoteOriginally posted by: paboHave you thought about using a genetic algoritm? These are quite good for high dimensional optimisation.