Using Poisson Markov Models to Predict Game States in StarCraft

Student thesis: Master thesis (including HD thesis)

  • Anders Hesselager-Olesen
4. term, Computer Science, Master (Master Programme)
This project aims to investigate the feasibility of using mixed observation
types in Hidden Markov Models, in an attampt to increase the accuracy
of recognising strategies, and predicting future actions in the domain of the
real-time strategy game StarCraft. The types of observations in the model
will be multinomial and Poisson distributions, and theory for both types of
variables will be presented, as will theory for the combination of the two.
The data for training the model in the StarCraft domain will be analysed to
determine the validity of applyig Poisson distributions to the production of
combat units. Finally, experiments will be made to determine the accuracy
of predictions made by the model, as well as an evaluation of the most likely
path through the state space of the model.
SpecialisationGame Programming
Publication date6 Mar 2014
Number of pages76
ID: 193316614