How many of you are Extreme Programmers?
Robin Becker
robin at jessikat.fsnet.co.uk
Fri Apr 18 01:23:33 CEST 2003
In article <b7n4b8$maq$1 at slb0.atl.mindspring.net>, Andrew Dalke
<adalke at mindspring.com> writes
>Robin Becker:
>> interestingly genetic programming et al seem to perform worse when
>> attempts are made to 'direct' the evolution.
>
>Pardon?
>
>I thought the point of the cost function was to 'direct' the
>evolution.
>
> Andrew
> dalke at dalkescientific.com
>
>
I think the *direction* P Hansen is talking here is in the solutions
that are selected; the programmers know how to choose the candidate
solutions.
In GP if you bias the choices then, almost always, the results are
inferior.
You are correct in saying that the cost function implicitly directs the
evolution; it does not determine how candidates are chosen. In simple
terms the candidates should be drawn from a set of non zero measure (ie
as large as possible in the set of all solutions). Conceptually GP is
direct search using an evolution like process for the candidate
selection so ideally we would like to cover all possible solutions, but
given the time constraint we cannot.
--
Robin Becker
More information about the Python-list
mailing list