Interesting Projects

datdota is a data-warehouse which serves as an archive for the history of professional Dota 2 games. It amalgamates data from various Valve WebAPIs, as well as extracted data from the protobuf-encoded replay files. There's also extensive hand-curation - making it the most extensive (56k+ matches) and accurate collection of Dota 2 pro data. It's coded in Groovy (as part of Grails 2 framework), Java (for replay parsing), and Python (with scikit-learn, for various classification and predictive enrichment models). Other technologies include PostgreSQL & clarity.

windrun is a website that track's players Ability Draft (a type of variation of normal Dota 2) game statistics. It uses a similar setup to datdota (except Grails 4). The replay-parsing is on a much larger scale, over 11 million replays (300TB+) have been parsed. It's the most visited Ability Draft specific stats website. coming soon.

The International Probabilities is a frontend for a simulation which models various outcome for The International. This attracts a lot of attention in the pre-TI craze as each game can shift millions of dollars in prize money. The underlying model is Monte Carlo, with individual matchups based on an ensemble of Elo & Glicko models. The backend is written in Python (with scikit-learn), and the frontend is with jQuery & Highcharts.