Oliver Schulte's Page On Sports Analytics Sports Analytics: Software and Datasets

Markov Game Model for the National Hockey League

Dataset. The complete NHL play-by-play data from 2001-2014. Provided by nhl.com. We manage the data inside a database system, so this is a dump from MySQL. Note: Before 2007, the data is not well standardized and does not include complete information about which players were on the ice at a given time.
Model. We built a Markov game model for the 2007-2014 play-by-play data. This includes the state transition graph and the Q-functions for goals and penalties. We manage the data inside a database system, so this is a dump from MySQL. Managing big models in an RDBMS works very well, this is the BayesStore design philosophy.
Analysis. Various tables that we use to evaluate the model, e.g. entropy conditional on states.
Java Code. Currently all the documentation we have is this readme file.

Citing the dataset. If you use the dataset or the model, please cite the following paper (BibTex).

Routley, K. & Schulte, O.

A Markov Game Model for Valuing Player Actions in Ice Hockey

Uncertainty in Artificial Intelligence (UAI), pp.782-791.

2015

Box Scores from the 2011-2012 season in Excel format. You can get these in properly normalized SQL format from the relational learning database repository.