❔Genetic Algorithms

Genetic algorithms are a machine learning technique that takes a group of neural networks that are trying to solve a problem and picks the ones that perform the best. Traits of the best performing neural networks are then combined randomly to form a new group of neural networks, this process is called β€œbreeding”. The process is then repeated over and over until you have a neural network that can perform a task well. Another version of the genetic algorithm is called Interactive Selection. The difference between this version and the original version is it allows people to decide which neural networks make it to the next generation. This is what we will try to achieve with breeding mounts and using them for betting. When a mount is minted it is given a random neural network that will be responsible for driving the chariot and using skills. As the mount is playing in the betting match the neural network will be in constant feedback to optimize the output given the sensor inputs in the match. As the mount runs in more and more races it will start to perform better and better. ( note the neural networks will only matter in the betting matches. PVP matches will not affect this) Since mounts will not be able to run an infinite amount of races it will be important for players to find smarter mounts and breed them together in order to get the best chances of winning a match. Neural network data will not be shown to the player so they will be forced to use match history statistics to tell how good any given mount is. Also since each skill will have its own neural network it will be important to choose the skill that the mount is best at. If the horse was a result of breeding then you will be able to see which horse the skill was inherited from. Looking at the horse that the skill came from you can see match history and win rate with that skill to determine if the horse will be intelligent with that skill. When breeding the skills and driving neural networks will be randomly selected from the parents with a 50% chance selection for each.

Last updated