[galib] a notion of speed in GALib - RFC
phhht at earthlink.net
Sat Jun 6 14:42:41 EDT 2009
For some time I've been thinking about how to get some measure, in
some sense, of how fast a population of genomes is changing.
The units of this measure seem apparent: bits/generation. Here's
what I did.
For concreteness, let's take GAInc. At the beginning of each step(),
I save the current state of a
genome. At the end of the step, I save the Hamming distance (in
bits, not percent) between the original and current states of the
genome. At statistics report time, I report the speed as the
arithmetic mean of the distances collected.
I'd like to hear from you with any comments you have. In particular,
is there a better or alternative way to measure speed? What's wrong
with this way?
This technique also works for GASimple. In GASState, the technique
is complicated by uncertainity
that I am measuring the initial and final states of the same genome.
But does that matter much?
Thanks very much for your comments.
More information about the galib