vardoom/flappygeneticsnn

Neural Network and Genetic Algorithm for Flappy Bird Game

Flappy Bird Genetic Evolution

This project aims to train an agent in the PyGame environment "Flappy Bird".

The training process uses a combination of Neural Network algorithms and Genetic algorithms.

Algorithm Process

• Each generation contains 50 birds
• At each generation, the breeding process is performed based on fitness scores
• Crossover swaps the first layers of the NN of the selected parents
• Finally, random mutations are performed in order to achieve diversity

How to use ?

The algorithm loads the trained models from the load_population folder. In order to do this, the LOAD_POP variable must be set to True in the utils.py file.

The algorithm saves the trained models on the saved_population folder. In order to do this, the SAVE_POP variable must be set to True in the utils.py file.

The number of generations and the size of teh population can be changed in the utils.py file.

Author

• Paul Vardon - Vardoom

Disclaimer

• Based on Flappy Bird clone in pygame, https://github.com/sourabhv/FlapPyBird
• Genetic algorithms inspired by https://github.com/ssusnic/Machine-Learning-Flappy-Bird