Bots and Simulations > Evolution and Internet Sharing Sims

Zerobot evolution prize for conditional logic

<< < (2/17) > >>

EricL:
Fine with me.  A vector of random numbers would also be okay.  Any set of starting numbers would be fine as long as the entropy does not differ significantly from that of a pure zerobot.

Sprotiel:

--- Quote from: EricL ---Fine with me.  A vector of random numbers would also be okay.  Any set of starting numbers would be fine as long as the entropy does not differ significantly from that of a pure zerobot.
--- End quote ---
Entropy is certainly not a relevant characteristics: a pure zerobot has minimal entropy, while a purely random genome has maximal entropy. Intuitively, both should be admissible because they incorporate no relevant information wrt. bot behaviour and evolution.

EricL:
Sorry, I should have said they must have similar relevant information entropy (otherwise known as Kullback–Leibler divergence) relative to the set of all "functional" bots.

You are corret to say the entropy of a random sequence is higher than a sequence of all zeros relative to the space of all possible seqeunces, but the set of all possible sequences is not the relevant set we care about.  A zerobot and a random bot both have similar information entropy relative to the set of all functional bots (bots that do something interesting including all hand authored bots).  Said another way, their information content is zero or nearly so relative to bots that do something interesting.

I think the root of the issue is that in a evolutionary system, the definition of "organized" is not the same as "statistically improbable".  A zerobot is statistically improbable but as far as DB is concerned, it is unorganized and thus exhibits high relevant entropy.

Endy:

--- Quote ---I can attest that a static environment with modest costs at the beginning isn't the way to go. I have a sim in the hundreds of millions of cycles and nothing noteworthy.
--- End quote ---

I've gotten decent results by mixing up the enviroments. First I started off with a gravity/brownian motion enviroment, with veggies giving free energy shots. As bots learned to feed/repro I installed standard veggies. As the bots adapted and learned to move to some extent, I lessened the amount of gravity and brownian motion in the sim.

Evolution seems to work best if it has something to work on. The bots need to have increasingly harder challenges to test their mettle against.

Numsgil:

--- Quote from: Endy ---Evolution seems to work best if it has something to work on. The bots need to have increasingly harder challenges to test their mettle against.
--- End quote ---

I'll second that.  Evolution doesn't like to make progress, you have to drag it kicking and screaming to the next plateau.

Navigation

[0] Message Index

[#] Next page

[*] Previous page