Thanks to MartinBraendli for his Volley (Neural Network) Thanks to SBissay for his Emergence and relative Forum (in franch) Take a look at the new version: https://scratch.mit.edu/projects/196203930/ The project has been largely rewritten in the AI part. Here are the main new features. The gene of the genome is now of two hexadecimal digits (not three). There is a new ‘train’ process based on a genetic algorithm. All the 100 genomes of the population play a game of N (slider variable) rallies. At the end genomes are sorted on a fitness function based on the number of rallies won by each attacker and the genetic algorithm follows as usual: the best fitted genomes are most likely to be selected as parents of new children and top ones are chosen as member of an ‘elite’ with the purpose not to be altered. There is also a club (circular list) of ‘super elites’ formed of genomes that won big a game. To increment diversity a certain number of ‘mutations’ are inserted. Then a new generation begins. Deserve some attention the 'breed' blocks: they are duplicated as to present a different approach to the problem and a certain variety of 'mutation' operators. It was also added e new item in the menu: ‘aivai’, you select an attacker from population and let him play against defender. All the project is a good introduction to Artificial Neural Network and Genetic Algorithm.
