[galib] Average score stagnates

John V Kew peri at logorrhea.com
Fri Apr 23 11:39:04 EDT 2004

I've got a 10 genome agent solving a simple parametric function. I'm
analysing sharing mechanisms between GAs, and this particular agent is
the only agent of ten which can share it's solutions with the larger
population of agents.

I have an odd occurance.

This agent will very quickly reach a sub optimal average score and never
move beyond that score. Even though the genomes in the agent appear to
be changing (They are relatively homogenous, but they are changing), the
average score never changes. 

What is more odd is that other agents, upon receiving solutions shared
by the sharing agent, are naturally drawn to this score but also begin
to stagnate.

The level of sharing is low enough (one solution shared every 100
generations) such that some modifcation would be expected. I'm using the
Bin2Dec genome and the default mutator. The mutation and cross over rate
is low, but not low enough to prevent optimization.

My question is:
What characteristics typically stagnate average scores in GAs using GaLib?

More information about the galib mailing list