Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!news.mathworks.com!uunet!in1.uu.net!world!rie
From: rie@world.std.com (Daniel Rie)
Subject: Re: Hamming distance for TSP?
Message-ID: <D6qH8F.85t@world.std.com>
Organization: The World Public Access UNIX, Brookline, MA
References: <D67pyC.2Mr@murdoch.acc.Virginia.EDU> <1995Apr6.191732.25072@wisipc.weizmann.ac.il>
Date: Sat, 8 Apr 1995 20:36:14 GMT
Lines: 24

oren@hadar.weizmann.ac.il (Ben-Kiki Oren) writes:

>Robert C. Craighurst (rcc6p@uvacs.cs.Virginia.EDU) wrote:
>> I want to get some kind of measure of the diversity of my GA population.
>> One approach is to us some sort of Hamming distance measure...

>...... I'd say that the simplest thing is to measure the entropy of the
>population. I assume you use some sort of a fixed-length representation, and
>that each position can hold one of a finite set of symbols. 
>Then measuring the Shanon entropy is trivial, fast, and illuminating.
Am I missing the obvious?  Let's say there are two symbols, 0 and 1.  Would
the entropy calculation or even a data compression algorythm ignore the
position of each symbol, and focus on repeated sequences?  Two very different
solutions could result in the same probabilities in the entropy calculation.
It might be more appropriate to use a directed divergence calculation which
can be equivalent to an entropy calculation, but is actually a distance 
measure. Entropy or directed divergence sounds right as a measure of 
diversity, but I'd like to get some specifics of how to constuct a good
measure for populations of problem solutions.
-- 

============================================================================
Dan Rie                             
Scituate, MA
