I'm running a velocity-experiment, with velocity as the only selection
criteria. Some nice symmetrical creatures have been evolved (but speed is
still bery bad).
Problem is, brain size is expanding too much. Most of it is just "lose"
neurons with no connections. This consumes a lot of processing power, and I
get between 15 and 30 warnings for most creatures "borned" :
LiveLibrary::create [WARN] 27 warnings while creating....
using CLI.
I've tried to "select away" the brain, but at least with my low speeds it's
impossible. Even with 10000 points for velocity and -1 for brain size,
during some days brain size equals zero since it's much easier for evolution
to strip the brain size than to improve velocity.
How can I get less mess in my brains?
(Yes, I do have a lot of neuron types active. I run this experiment as
complicated as I can, with feeding, fighting, water/ground, heigt-field etc.
Velocity will be replaced by Distance later on when speed is present. Goal
is to start from scracth with as complicated (real-worl-like) experiment
definitions as possible and run for as long as possible, to see if I can
create robost creatures that can survive in many environments)
Tjalve
Re: How to get rid of Waste Brain Neurons
> Problem is, brain size is expanding too much. Most of it is just "lose"
> neurons with no connections. This consumes a lot of processing power, and I
> get between 15 and 30 warnings for most creatures "borned" :
> LiveLibrary::create [WARN] 27 warnings while creating....
> using CLI.
These warnings are usually caused by neurons which cannot work
together in the same place of a creature (for example, many
muscles of the same type in the same place). See the "error
reporting" settings to get rid of this warning, or to see
their detailed report.
> How can I get less mess in my brains?
You can add a criterion for the fitness function, like
brain size, with a negative weight (and perhaps turn on
normalization).
Note that this might somehow prevent evolution from progress
in brain, as a creature with no redundant neurons will always
be better than the same creature with one or more
redundant neurons. It depends on the values of weights,
gene pool size, and other parameters like selection
and deletion.
Yours,
Maciej