This repository contains supporting code for our paper A Metric Learning Approach to Anomaly Detection in Video Games. All of the models used in experiments can be found here along with an interactive demonstration (does not require dependancies and is run online). Models are written in pytorch.
The atari-anomaly.ipynb file is the main file used to train and visualise some aspects of the models.
This repo requires some github dependancies, to avoid backward compatibility issues it is best to use the particular commits given.
- pyworld
git clone -n git@github.com:BenedictWilkinsAI/pyworld-rl.git
git checkout -b master a15675e9b8fa333312e73e751bb737fd4649d3b5
- anomapy
git clone -n git@github.com:BenedictWilkinsAI/anomapy.git
git checkout -b master 11f02ca42c375fc9c39875ba072541c1d8bf9bea
- dataset
git clone -n git@github.com:BenedictWilkinsAI/pydatasets.git
git checkout -b master a9a6fe479361844b57fe662768a6489353ac86b8
Our dataset (AAD) was used to evaluate the performance of S3N, the dataset and more information can be found here.