Analysing Openings in Tactical Simulations

‘Openings’ are strategies for playing the beginning of a game, and are often described as sequences of low-level actions (as in chess, for example). This makes it easy to query a database of recorded games for useful information, such as an opening’s success rate, popularity, and more. ‘Tactical shooters’ are video games featuring pseudo-realistic tactical simulations. Here, openings are conventionally described using natural language, which is general enough to capture an opening’s individual variations but provides no formal means of recognising it. Using Valve’s Counter-Strike: Source as a research platform, this thesis describes the use of cluster analysis over a database of recorded games to formally distinguish between openings in tactical simulations. Empirical results reveal significant correlations between openings and game outcomes. The same process is also shown to effectively model player behaviour, and has potential future application to player classification and the development of believable computer-controlled players.