15.12.2012 Views

Bayesian Programming and Learning for Multi-Player Video Games ...

Bayesian Programming and Learning for Multi-Player Video Games ...

Bayesian Programming and Learning for Multi-Player Video Games ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

7.8 Conclusion<br />

We contributed a probabilistic model to be able to compute the distribution over openings<br />

(strategies) of the opponent in a RTS game from partial <strong>and</strong> noisy observations. The bot can<br />

adapt to the opponent’s strategy as it predicts the opening with 63 − 68% of recognition rate<br />

at 5 minutes <strong>and</strong> > 70% of recognition rate at 10 minutes (up to 94%), while having strong<br />

robustness to noise (> 50% recognition rate with 50% missing observations). It can be used in<br />

production due to its low CPU (<strong>and</strong> memory) footprint.<br />

We also contributed a semi-supervised method to label RTS game logs (replays) with openings<br />

(strategies). Both our implementations are free software <strong>and</strong> can be found online 10 . We use<br />

this model in our StarCraft AI competition entry bot as it enables it to deal with the incomplete<br />

knowledge gathered from scouting.<br />

We presented a probabilistic model inferring the best army composition given what was<br />

previously seen (from replays, or previous games), integrating adaptation to the opponent with<br />

other constraints (tactics). One of the main advantages of this approach is to be able to deal<br />

natively with incomplete in<strong>for</strong>mation, due to player’s intentions, <strong>and</strong> to the fog of war in RTS.<br />

The army composition dimensionality reduction (clustering) can be applied to any game <strong>and</strong><br />

coupled with other techniques, <strong>for</strong> instance <strong>for</strong> situation assessment in case-based planning.<br />

The results in battle outcome prediction (from few in<strong>for</strong>mation) shows its situation assessment<br />

potential.<br />

10 https://github.com/SnippyHolloW/OpeningTech/<br />

156

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!