Global State Evaluation in StarCraft
Graham Erickson, Michael Buro · Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment · 2014
State evaluation and opponent modelling are important areasto consider when designing game-playing Artificial Intelligence.This paper presents a model for predicting whichplayer will win in the real-time strategy game StarCraft.Model weights are learned from replays using logistic regression.We also present some metrics for estimating player skillwhich can be used a features in the predictive model, includingusing a battle simulation as a baseline to compare playerperformance against.