Re: distributed processing ad multiThreading
Re: distributed processing ad multiThreading
- Subject: Re: distributed processing ad multiThreading
- From: Robert S Goldsmith <email@hidden>
- Date: Mon, 17 Sep 2001 16:52:17 +0100
Hi :)
I'd consider running parallel distributed simulations (for perhaps
hundreds or thousands of generations) and then 'cross-populate'
between those simulations. This should reduce the overhead
introduced by the distribution mechanism itself, and has the added
side effect in many simulations of reducing the impact of local
maxima being reached in one or more simulations, and generally
speeding up the reaching of optimal solutions. This side-effect
might well be useful in your case where you're running small
populations of (presumably complex) entities.
This is one of a few methods I would love to use. Another
method I am working on myself and have already shown to give
much better results than some base generational code is
using steady state selection (only replace one entity at a
time - not the entire generation) coupled with a local
tournament selection. Again, it gives local neighbourhoods
and is much better at improving the general fitness level
instead of steaming ahead with one or two super-fit
entities. However, for comparative study, I have to use the
methods other people have published results on. This means
standard gp, generational, elitist selection and generally
cr*p performance. In some ways this is good because the new
methods I am working with really are very very much better
than these base level methods :)
It does, however, mean i am stuck with the problem I
described before :(
In systems where the return time on calculations is greater, I
can see easy ways to distribute. For instance, my main research
area is in evolutionary hardware and this will need parallel
running of about 6 very complex, processor eating neural
networks which I can distribute easily because they will run
for over 15 mins almost independently.
The crux of the current problem (for a paper I am working on
for a conference next year) is that the base, bog standard
gp is not easy to thread or distribute.
I was hoping something ObjC would make life a little easier :)
Robert