Mutation rates in evolutionary corewar

[Guest]

Archived from groups: rec.games.corewar (More info?)

I have finally figured out why all my evolved nano warriors look mostly
identical: The mutation rate is too high.

The problem is when the mutation rate is too high, the warriors have
significant pressure to assume a mutation resistant form. Going through
the results of the new evolver showed me that all the typical forms I'd
evolve from the previous version, were terrifically resistant to some
twiddling or its bits. On a hunch, I dropped the mutation rate by a factor
of about 7, and I suddenly started evolving more interesting forms that
scored better, but did not reproduce effectively in the face of high
mutation rates.

This effect seems to be most pronounced on nano, because 5 line warriors
can change form in a surprisingly short amount of time. This means that if
the mutation rate ever climbs too high, all the complexity evolved is
effectly lost as the warriors assume resistant forms.

Something to think about.

 

[Guest]

Archived from groups: rec.games.corewar (More info?)

Interesting point.
I"ve seen this. High disruption forces the GA towards short, simple
forms such that mutation or crossover doesn't change them so much. Less
disruption sometimes allows more sophisticated warriors to develop
which leads, alas, to weaker performance in my experience anyway Oh
the perversity of it all!

In the same vein: coevolving warriors sometimes seem to evolve to be
better at evolving. I'll get warriors where a single point mutation
turns replication on or off or switches core clearing and imp bombing
back and forth.
Dave Hillis