Training a neural network-based snake game agent using genetic algorithms for evolutionary improvement
Every agent on the screen has own neural network.
During the initialization step of population creation, each agent is assigned unique weights.
Eventually:
- These unique weighted agents will evaluate by their fitness functions, and selected by genetic algorithms.
- Every generation or epoch after the first epoch, there will be choosen agents who shows successes about enclosing to apple.
- The best agents consist of areas that show the closest moves to the apple