Flappy Bird Game : trained using Genetic algorithm

Introduction

---

FlapAI is a genetic algorithm coded in Python teaching itself how to play Flappy Bird. It can reach score of 1000+ pipes in less than 100 generations with population of just 100. It is based on FlapPyBird. All settings can be found in config.py and can be adjusted accordingly. Neural networks are saved in a json file for further evaluation. Each time you run FlapAI, statistics and the neural network of the best performing bird are saved in the directory save/

Features

---

• Customizable genetic algorithm
• Speed up (FPS can be set)
• Statistics
• Running a single Neural Network from a json file
• Cool ASCII art
• Neural Network - It is 4 layer NN with customizable number of neurons in each hidden layer.It is trained using genetic algorithm with population of 100.

Requirements

---

• pygame
• numpy
• matplotlib
• colorama (cool colored print)

Usage

---

If you want to train birds with the parameters in config.py just run FlapAI.

python flapai.py

To evaluate a single Neural Network use -evaluate (some can be found in /interestingannsave)

python flapai.py -evaluate neuralnetwork.json

If you want to see the statistics of a past experiment use -stats

python flapai.py -stats save/2016-08-27_17:16:04

Keyboard commands

---

Press UP ARROW while focusing the game window to speed up the game (if the screen is frozen it means that pygame doesn't redraw the window anymore which speeds up the algorithm a lot)

Press DOWN ARROW while focusing the game window to slow down the game

Press ESC while focusing the game window to stop the algorithm and show the statistics (don't use CTRL+C in the terminal)

Screenshots

---