codeandkey/coms527-project-rlc4

a distributed reinforcement learning agent for Connect-4

Justin Stanley - COMS527 Final Project

A distributed reinforcement learning system targeting Connect-4.

Starting

This project requires PyTorch 1.2.1+, numpy, and a computer with a GPU available. The learning hyperparameters are in drl/param.py.

To start the learning process,

$ mpirun python drl.py

To view the current training progress,

$ python results.py

A window showing the loss over time and model performance will appear, and a file results.png will be written with the same information.

Systems Architecture

Model Architecture