Pong AI

An rallying AI that plays Pong

By Sam Reiter, Adie Guo, and Sam Robinson

Full video

Dependencies

• Python
• numpy
• pandas
• matplotlib
• pygame

To install:

With pip:

python -m pip install numpy pandas matplotlib pygame

With pacman:

sudo pacman -S python-numpy python-pandas python-matplotlib python-pygame

With conda

conda install --Dependencies

Usage

Run

python train.py

If you'd like to play pong by yourself run

python pong.py

File Descr.

agent.py

Our RL agent which interacts with the game

functions.py

Contains our loss function and activation functions

layers.py

Layers of our model

network.py

Our network of layers

pong.py

Code for the game pong

train.py

run to train the model

Dependencies

dependencies and version numbers we used

References Used

This project was coded without machine learning libraries. Here is a playlist of videos used to achieve this.

Next Steps

• make the bias work
• train after each move rather than each game
• Integrate a proper reinforcement-learning algorithm.
• Add a second AI to play against the first.
• Use GPU acceleration.
• Create binaries for different OSs so dependencies aren't needed