PureBytes Links
Trading Reference Links
|
Samerski, Michael (Australia) writes:
> I want the optimiser to use a more efficient algorithm than a linear
> search to find the optimum system parameters.
The problem here is that if a system performs great with a parameter of
3 and pretty good with a value of 5, you can't draw any conclusions
about 4. You just have to try it, so eventually, every parameter would
need to be tried anyway.
And as a human, trying to tweak something, would you be satisfied if a
system tester came back and said "values 1, 3, and 5 work, but I'm not
going to bother trying 2 and 4"? I wouldn't; I want them tried. As as
far as stepping through a series of values, it doesn't get any more
efficient than incrementing that value.
> I suggest using a Genetic Algorithm or Neural Network.
I'd argue that these are no more efficient in terms of time OR space.
> Currently the linear search (stepping through all values in a range)
> is hopeless for small systems with only 2 parameters (500 tests) on a
> P200.
I'd suggest buying a little more horsepower.
--
========================================================================
Alleged medical quotes: *She slipped on the ice and apparently her legs
went in separate directions in early December.
http://www.unixgeek.com/cgi-bin/motd.pl - PGP email preferred
|