OwlZero

Project: OwlZero

OwlZero is a machine learning AI applied to the game of Gothello.

This AI is a Monte Carlo Tree Searcher, that learns by playing games and figuring out which moves are most likely to lead to a victory for the player.

OwlZero plays against Grossthello, a Depth Limited Negamax AI.

OwlZero's top win rate is 96.0% over 150 games at 300 rollouts.

Gothello

Gothello is a combination of Go and Othello played on a 5x5 board.

Each player takes a turn by placing a black or white stone on the board and captures an opposing group if they completely surround them.

You may not place a stone in an area that would cause you to be surrounded or form a group with no places to move.

Intelligent Gameplay

We can see OwlZero improving its own ability to play Gothello as it plays games.

First Game

OwlZero tries to divide the board horizontally but ends up needing to surround Grossthello to pull off a narrow win.

Seventh Game

OwlZero has learned that dividing the board diagonally using a checkered approach is much more effective for controlling large areas of the board.

61st Game

OwlZero has learned how to control large areas of the board using checked piece placement and diagonal board division.

OwlZero learns by simulating, playing and building routes to winning games with rewards in a 4 step process across all the games it plays.

Selection: Find the route that is most likely to win from the current state

Expansion: Expand our route with new possible game states

Simulation: Simulate possible new games from that expansion

Backpropagation: Updating our route with the rewards we found

OwlZero Performance

Across the board we see OwlZero winning an average of 93% of all rounds played versus Grossthello.

100 Rollouts - 93.4%
200 Rollouts - 89.8%
300 Rollouts - 96.0%

If we looking at runtime performance we get the most value from running OwlZero at 100 Rollouts.

Conclusion

From the above results we can see that Gothello is very receptive to Monte Carlo Tree Search machine learning. We noticed through the learning process that OwlZero reaches its maximum effectiveness very quickly so we can conclude that mastering the game of Gothello is a fairly quick process.

Report abuse