Mohammadreza Daliri wrote:OpenSpiel is an open-source software system for implementing high-performance software players for many different computer games. Hex is a two-player game of perfect information used in a variety of computer games research projects. The OpenSpiel project has implemented a version of the AlphaZero algorithm used to great success in the games of Go and chess. Therefore, we use an existing implementation of Hex in OpenSpiel, along with the AlphaZero algorithm, to train and evaluate strong Hex players.
We perform experiments with OpenSpiel for both 6 by 6 and 8 by 8 Hex where, to improve the diversity and coverage of self-play games during training, a set of so-called forced-move openings are used. We show that the forced-move openings allow OpenSpiel agents to learn faster, as measured by a test suite of positions. We also developed a set of test positions, with the help of the MoHex system, which can prove the game theoretic value of a given position. Head-to-head play of game agents is often used to evaluate player strength, but a test suite based on proven winning moves (via MoHex) is computationally less expensive.
Implementing and improving the AlphaZero algorithm
1 post • Page 1 of 1
https://era.library.ualberta.ca/items/7 ... ee2e0ccaa1