optimization algorithm

[Muddy]

I need an API that would allow me to find the inputs to any function that minimize (or maximize) that particular function. It would have to be used as follows:

1) initial function inputs are a guess (this is my software)
2) function is calculated (this is my software)
3) result from #2 above is fed to optimization engine that suggests the next set of inputs to try (this is the software I need)
4) loop back to #2 above. (this is my software)


Does anyone know of any vendors for such software?

[Jared]

Well, I am just throwing some ideas in.... maybe it will take you somewhere, maybe not.....

Instead of Vendor or Third Party solutions, implement your own:

If you're domain, or input into the function is known, and is not to incredibly large, you could start at the lowest value X which gives the function's output Y. Depending on the size of your domain, if it is rather small, and the computation won't take to long, run all the X's through your function, then take the max/min of Y. If you're domain is large, then you could use algebraic methods to find "turning points", then use that as narrowed domains, to home in on all local max and mins, then compare those.

Anyway, just some abstract ways at approaching the problem, as I understand it.....

[Muddy]

actually the domain is huge, the function isnt continuous, and there are many local maximums and local minimums (i want the global minimum of course) making this a nasty problem better suited to those with abilities far greater than mine. This is the reason I am looking for a third party solution.

Thanks for the input though!

[Shaggy Hiker]

Actually, this doesn't sound all that difficult. Look into a steepest climb algorithm. I would suggest an evolutionary algorithm, but from what you said, I thought the steepest climb might get you there faster, you've almost done it already.

Guessed at set of inputs
alter each one in turn looking to see whether you move closer to your goal.
Take the one with the most change as your new set of inputs.
Repeat until no further optimization can be done.

Is that along the right lines?

[Matt_T_hat]

Perhaps I have misunderstood the problem but would not a method not disimiler to a binary chop work well.

For those who don't know the binary chop (shame on you as it is very fast) you find the mid point of the range and ask higher(T/F) given 22K possibilities you now have 11K, 5.5K, 2.75K, in just 3 tries we have illiminated 19250 of the 22000 possibilities. repeat untill fully satisfied.