Improving AI in Skat through Human Imitation and Policy Based Inference

• Author / Creator

  Rebstock, Douglas

• Creating strong AI systems for trick-taking card games is challenging. This is mostly due to the long action sequences and extremely large information sets common in this type of game. Thus far, search-based methods have shown to be most effective in this domain.

  In this thesis, I explore learning model-free policies for Skat, a popular three player German trick-taking card game. The policies are parametrized using deep neural networks (DNNs) trained from human game data. I produce a new state-of-the-art system for bidding and game declaration by introducing methods to a) directly vary the aggressiveness of the bidder and b) declare games based on expected value while mitigating issues with rarely observed state-action pairs. While bidding and declaration were improved, the cardplay policy performs marginally worse than the search-based method, but runs orders of magnitude faster.

  I also introduce the Policy Based Inference (PI) algorithm that uses the resultant model-free policies to estimate the reach probability of a given state. I show that this method vastly improves the inference as compared to previ- ous work, and improves the performance of the current state-of-the-art search based method for cardplay.

• Subjects / Keywords

  • Inference
  • Search
  • Skat

• Graduation date

  Fall 2019

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-mxx7-s832

• License

  Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.