A final project in an undergraduate machine learning class. It is a simple win/loss predictor that uses a hinge loss classifier to try to predict the outcome of a match before it has started. The highest accuracy achieved was about 60%.
This project also contains code for gathering data from the Steam API and compiling it into a more useful CSV format. To compile hero winrate information, the corresponding DOTABUFF page needs to be downloaded and saved as an HTML file before running the program. Refer to the code for exact file locations and naming.