This project aims to play Wikiracer using the text embeddings of the articles on Simple English Wikipedia. It can train models for link prediction and distance estimation, and it can run the distance estimation models on A*.
data: Contains the downloaded raw data from The Wikimedia Foundation, as well as the MySQL and Memgraph database files. Also contains the intermediate data generated for training and evaluation.prep: Contains scripts that were used while preparing the data for training, such as for retrieving the text embeddings from OpenAI or computing the PageRank of every node.eda: Contains scripts for analyzing the distribution of distances and PageRanks on the graph, as well as scripts for characterizing the text embeddings.link-prediction: Contains scripts to train and evaluate logistic regression models for link prediction.distance-estimation: Same as above but for distance estimation.graph-search: Contains code for running A* with the heuristics generated above or with the "Null" heuristic.docker-compose.yml: Has profiles for Memgraph and for MySQL, pulling files from thedatadirectory.config.yaml: Has paths to all the data files.