|Project Name||Stars||Downloads||Repos Using This||Packages Using This||Most Recent Commit||Total Releases||Latest Release||Open Issues||License||Language|
|Program learning to play Flappy Bird by machine learning (Neuroevolution)|
|Deep Neuroevolution||1,502||2 years ago||17||other||Python|
|Machine Learning for Flappy Bird using Neural Network and Genetic Algorithm|
|Evolutionary Algorithm||1,092||2 months ago||4||mit||Python|
|Evolutionary Algorithm using Python, 莫烦Python 中文AI教学|
|Gpumd||200||2 days ago||12||gpl-3.0||Cuda|
|Graphics Processing Units Molecular Dynamics|
|A neuroevolution game experiment.|
|Darwin||91||a year ago||1||December 08, 2020||apache-2.0||C++|
|Evolutionary Algorithms Framework|
|Super Mario Neat||74||8 months ago||1||mit||Python|
|This program evolves an AI using the NEAT algorithm to play Super Mario Bros.|
The goal of the project is to apply NEAT to evolve the network architecture of Deep Neural Networks for improving their performance on a given classification task, in this case the game Flappy Bird, the goal is to determine if for a given input the bird must jump or no to avoid hitting the pipes. This repository contains a Jupyter Notebook where is possible to compare a Human, a Neat implementation and a multilayer perceptron playing Flappy Bird.
For both, neuroevolution algorithm and multilayer perceptron the input is the same:
1) Distance in the X axis from the bird to the next pipe. 2) Distance in the Y axis form the bird to the lowest point of the pipe in the top. 3) Distance in the Y axis from the bird to the highest point of the pipe in the top 4) Distance in the Y axis from the bird to the top of the map. 5) Distance in the Y axis from the bird to the bottom of the map.
In this project we probe that NEAT can improve the performance of a DNN. We found that NEAT can be useful to reduce the complexity of our models. NEAT is able to produce networks that produce similar results to the MLP but much less complex. We have been able to use NEAT to reduce the complexity of our model. We went from 20 neurons and 98 weights in the MLP to 7-8 neurons and 6-10 weights, with is a huge complexity reduction.
More information available in the documentation.
Iker García Ferrero - ikergarcia1996 Gonzalo Pierola - Guamedo
To run this project you need Python and Jupyter Notebook. Also you will need to install all the python libraries specified in the "README.md" inside the "Jupyter Notebook" folder.
Documentation is available in the releases section with the corresponding release.
The implementation of the Flappy Bird Game that we used for this project is based on this code: Gamedevlapse: Create Flappy Bird in Python [Time Lapse] (youtube)