Training AI model to play Flappy Bird

. Tools used: Pygame(for developing the game interface), Python NEAT (Neuro Evolution of Augmenting Topologies)

NEAT Algorithm Process:

1. Inputs: Bird Y (Position of bird on y axis), Top Pipe (position of Top Pipe), Bottom Pipe (position of Bottom Pipe)
2. Output : Whether to make the bird jump or not jump
3. Activation Function : TanH
4. Population Size : 10 Birds (can take any larger number, but taking larger size will overfit the model, as it's a very small game)
5. Fitness Function : The fitness of the Bird is dependent upon how long it plays the game (how far it can go)
6. Maximum Generations : 30, After running 30 Generations, if the Algorithm couldn't crack the game, then it means it failed(in our test case)

How the birds are trained in each generation? :

In 1st Gen, 10 random birds will be played, then the best among them will be choosen and mutated and populated with the next generation birds.