Post by alandau on Aug 25, 2017 21:41:52 GMT -8
I've posted about this elsewhere when the first version of the charts went up (and I won't repeat it all here): alloyggp.blogspot.com/2017/03/uct-charts.html
Basically, UCT charts are a visualization of how a UCT algorithm fares when playing a given game. It shows how much an increased number of iterations of the UCT algorithm contributes to its chances of winning, and how much the game is skewed towards one player or another (as revealed by this statistic).
Since the original writeup, I've put together a fully open-source version of the code generating the charts (not yet documented, unfortunately -- I'll get around to that eventually): github.com/AlexLandau/game-ai-research
The new generation of generated charts will be posted at: alexlandau.github.io/alloybox/reports/testuctchartexperiment.html
I'm storing more data from the matches this time around, so I'll be able to pose questions like the preferred opening move.
Basically, UCT charts are a visualization of how a UCT algorithm fares when playing a given game. It shows how much an increased number of iterations of the UCT algorithm contributes to its chances of winning, and how much the game is skewed towards one player or another (as revealed by this statistic).
Since the original writeup, I've put together a fully open-source version of the code generating the charts (not yet documented, unfortunately -- I'll get around to that eventually): github.com/AlexLandau/game-ai-research
The new generation of generated charts will be posted at: alexlandau.github.io/alloybox/reports/testuctchartexperiment.html
I'm storing more data from the matches this time around, so I'll be able to pose questions like the preferred opening move.