Amazing video mate! Would be lovely to see your implementation of this in action. I think there's a mistake at around 11 minutes. You talk about the importance of the connection ID to identify similar connections, but then you say that a new connection between the same neurons that arise in a new mutation will be given a new ID and not the same. At first I thought that maybe there's an idea behind this to differentiate between evolutionary paths but this would cause issues with crossover if the same connection have several IDs.
@chrismarriott-CS Жыл бұрын
This is a feature of the original algorithm but has probably been adjusted on one of the many variants. The reason was to avoid having to look back at all connections to see if we have a match. Naively this is expensive. My next video in this series is a demo and it has a link to one of our live versions of you want to check it out.
@Dalroc Жыл бұрын
@@chrismarriott-CS That's interesting. I've interpreted all implementations as tracking the connections by innovation number that's based on the source and target node. With this interpretation there could potentially be multiple connections between the same two nodes? Or do you only allow a new connection if one doesn't exist in the current network? I saw the next video and it was very impressive and interesting. Quite similar to my own ideas. I wish you would've shown a bit more "under the hood" so to speak heh. It's really hard to wrap your head around each detail in NEAT.
@pratyushkhatua7123 ай бұрын
could I have the figures in this explanation? Do you mind if I use these figures in a presentation?
@chrismarriott-CS3 ай бұрын
Send me an email at dr.chris.marriott@gmail.com and I'll send them too you.