Labelling Multivariate Time Series Representation of StarCraft II Replays Using Hidden Markov Models

  • Gauthier Gerard

Student thesis: Master typesMaster in Computer Science Professional focus in Data Science


In recent years, the video game field has been used a lot for testing machine learning algorithms but none of them specifically targeted the classification of replays of video games. This is why this master thesis is oriented towards the automatic annotation of replays using HMMs.
Several formats of replay files exist, and the format that is used during this master thesis is one using multivariate times series in order to represent the game’s state at any given moment of a replay.
The usability of HMMs for the replay annotation problem has been evaluated using three experiments and the results are that HMMs could be useful for some annotations but more experiments have to be conducted in order to have a final word.
Date of Award23 Jun 2020
Original languageEnglish
Awarding Institution
  • University of Namur
SupervisorBenoit Frenay (Supervisor)


  • replay annotation
  • HMMs
  • semi supervised HMMs
  • time series

Cite this