A Bayesian model for opening prediction in RTS games with application to StarCraft

Abstract

Abstract-This paper presents a Bayesian model to predict the opening (first strategy) of opponents in real-time strategy (RTS) games. Our model is general enough to be applied to any RTS game with the canonical gameplay of gathering resources to extend a technology tree and produce military units and we applied it to StarCraft 1 . This model can also predict the possible technology trees of the opponent, but we will focus on openings here. The parameters of this model are learned from replays (game logs), labeled with openings. We present a semisupervised method of labeling replays with the expectationmaximization algorithm and key features, then we use these labels to learn our parameters and benchmark our method with cross-validation. Uses of such a model range from a commentary assistant (for competitive games) to a core component of a dynamic RTS bot/AI, as it will be part of our StarCraft AI competition entry bot.

Special tactics: A Bayesian approach to tactical decision-making

“…A Bayesian model for micro-management [3], in which units are attracted or repulsed by dynamic (goal, units, damages) and static (terrain) influence maps, actually moves the units in StarCraft. Other previous works on strategy prediction [4], [5] allows us to infer the enemy tech tree and strategies from incomplete information (due to the fog of war). …”

“…• [5], [4], in decision-making, we know our own possibilities (we know our tech tree as well as the units we own).…”

A Survey of Real-Time Strategy Game AI Research and Competition in StarCraft

“…These competitions, typically co-located with scientific conferences, have been possible thanks to the existence of the Brood War Application Programming Interface (BWAPI) 12 , which enables replacing the human player interface with C++ code. 14 . 26 total submissions were received from around the world.…”

“…No bot is capable of observing the opponent and autonomously synthesize a good plan from scratch to counter the opponent strategy. In this section we analyzed this claim quantitatively using the tools of [14]. We analyzed the replays from the 2011 and 2012 AIIDE competitions (shown in Figures 9 and 10 respectively).…”

“…Dereszynski et al [13] used Hidden Markov Models (HMM) to learn the transition probabilities of sequences of building construction orders and kept the most probable ones to produce probabilistic behavior models (in StarCraft). Synnaeve and Bessière [14] used the dataset of [4] and presented a Bayesian semi-supervised model to learn from replays and predict openings (early game strategies) from StarCraft replays. The openings are labeled by EM clustering considering appropriate features.…”