search

GMUEClab / ecj

ECJ Evolutionary Computation Toolkit
http://cs.gmu.edu/~eclab/projects/ecj/
122 stars 42 forks source link

NEAT Behavior #47

Open SigmaX opened 6 years ago

SigmaX commented 6 years ago

Jorge writes

if you look at the best-of-generation fitness plot, you can see that it jumps around a lot. At generation 13, for example, you have a best fitness of 4.0, while in generation you had 8.3. I've never seen this kind of behaviour with NEAT with a deterministic fitness function. Since the selection is elitist (top X% of each species), I think this should not be happening. The best-fitness should have a monotonous increase during evolution.

To me, this kind of behaviour suggests there is some genetic operation that is screwing up the networks sometimes, which could even explain both problems.

He observed this behavior when executing the following:

java -cp "." ec.Evolve -file ec/app/xor/xor.params -p stat.num-children=1 -p stat.child.0=ec.neat.NEATStatistics -p stat.child.0.file=neat.stat -p seed.0=199573726
SigmaX commented 2 years ago

I recall some back and forth about this, but I'm not sure if it was ever resolved.