Hi,
Assuming that the width of the range in more significant than
the resolution, you might try something like this:
heat = optimize(
"heat", 5, 1, 10, 1 );
fast = optimize( "fast", 25, 1, 50, 1 ) * 2;
slow
= optimize( "slow", 50, 25, 75, 1 ) * 2;
25,000 vs 100,000
combinations.
I've had good success with CMAE, you might try it
out.
OptimizerSetEngine("cmae");
It's pretty much
self-configuring, so don't worry about the parameters. If you are unfamiliar
with CMAE, there's a fair amount of good info in Help.
--- In amibroker@xxxxxxxxxps.com,
"Markus Witzler" <funnybiz@xx.> wrote:
>
>
Hello,
>
> if I have, say, three variables that I want to
optimize (exhaustively) where two have a range of 100 values and one would
have a range of 10 values, this would mean
>
> 10 * 100 * 100 =
100,000 combinations
>
> I figured that if I optimized the latter
two while keeping the first one fixed, that would take 10,000
combinations.
>
> Afterwards, I could use the optimal parameter
set for the last two ones and optimize for the first variable, i.e. 10
steps.
>
> Altogether, this would mean 10,100 steps as oppsoed to
100,000 steps.
>
> I understand that this procedure is not always
feasible. But in a case where one had for instance, a two MA crossover system
(100 steps for each MA) plus a heat parameter (10 steps), I guess this would
work.
>
> My reasoning would be optimizing for heat AFTER having
found the "best" parameter set regarding the MA´s would give me the highest
return (or else) without the need to run thru all theoretically possible
combos.
>
> Any thoughts on this besides using intelligent
optimization algorithm?
>
> I´m at a point where exhaustive
optimization is taking quite a while but still would be an option if I could
somewhat decrease the number of theoretical steps.
>
> Of course
with a larger number of opt. steps, intelligent optimization (using IO) would
be the ONLY option (I´m using IO anyways but am eager to find THE best and
most robust set of variables in the system I´m observing...).
>
> Any thoughts?
>
> Thanks
>
>
Markus
>