Randomness as a parameter to total fitness

[portaloffreedom]

Does it makes sense to have a increasing randomness in respect to the total fitness of previous saved evaluations?

It means that it will increase the randomess if the old evaluation get worse. With aging of the old evaluations' fitness, finding no new solutions will result in the old ones aging and therefore greater randomness on new walking patterns.

[silvanx]

Do you mean that you want to keep track of sum of all fitness values in the ranking and, if the sum doesn't increase with time, increase sigma in the random part of RLPoWER (that is now steadily decreasing with time)?

Or are you gathering ideas for a new learning algorithm?