The Starcraft Scouting Dataset

This dataset contains log data from 509 professional-level Starcraft: Broodwar games. All of the games are Protoss versus Terran matchups, and the dataset focuses on predicting the Protoss player's choices from the Terran player's perspective.

The games are divided into 30-second time steps, and the first 14 time steps from each game are retained. For each game, the data includes

• The total number of units of each type that were observable in each time step
• How many units of each type were produced
• How many units of each type were seen
• The Terran player's "scouting effort", which is the percentage of the Protoss player's Main Base + Natural Expansion that the Terran player could see

The README file included with the dataset explains the contents of each file in more detail.

All uses of this data should cite the following paper:

• Hostetler, J., Dereszynski, E., Dietterich, T., and Fern, A. (2012). Inferring strategies from limited reconnaissance in real-time strategy games. Proc. 28th Conference on Uncertainty in Artificial Intelligence (UAI 2012) (to appear).