Hello,
We're working on a model with sensory evolution. There are five possible
scents in the world. The predator, prey and food populations each start out
with their own scent, this can change during evolution. The predator and
prey populations start out with ten scent neurons, two for each scent. How
strong a creature's smell is is represented in user field 3, it's a vector
with values for each possible scent.
The problem, now, is that running it is very slow. Each scent neuron has to
run through all user field 3 vectors from the other creatures to calculate
it's state. With 30 creatures we can only run 4 steps/second on a pentium4.
Does anyone have ideas about how we can speed this up? Can we represent the
scents in another way that might speed up the simulation?
Thanks in advance,
Joost Wegman
Virtual life lab, Utrecht university
Re: Sensor complexity
> The problem, now, is that running it is very slow. Each scent neuron has to
> run through all user field 3 vectors from the other creatures to calculate
> it's state. With 30 creatures we can only run 4 steps/second on a pentium4.
> Does anyone have ideas about how we can speed this up? Can we represent the
> scents in another way that might speed up the simulation?
A good (technical) news is that in v2.9rc3, in FramScript,
senseCreaturesProperty can be used for Vectors (eg. "user1[7]").
It will speed things up. The v2.9 release is expected next week.
However, as you mentioned, due to the compexity of n^2
(each one has to smell all others), another idea may be required.
There are a few possibilities to choose from, we will see whether
the 2.9rc3 speedup will suffice or not.
MacKo