Post by Steve Draper on Feb 4, 2014 15:35:07 GMT -8
Just an interesting (and perhaps blindingly obvious with hindsight) observation about the difficulty (Agon) ratings for game (roles) that Tiltyard generates from statistical measures - regardless of the number of matches played (so even asymptotically) the values it converges to are HIGHLY dependent on the population of players.
The game that brought this home to me was 'Sheep and Wolf'. This is a strong win for the sheep, but it's also fairly easy for the sheep to make a critical error and lose. The net result is that currently the sheep are considered FAR the harder role on Tiltyard, yet a sufficiently competent player (several Tiltyard players are now) will essentially always win as the sheep. The equilibrium values with the player population seen up to now are strongly ANTI-correlated with the values it will converge to in the relatively near future (assuming players continue to generally improve). Right now the player population is probably very close to the tipping point for this game.
It's fairly probably that the larger connect 4 variants exhibit the same anti-correlation (to a much lesser degree) - at least if those board sizes are actually black wins. Obviously many of the games are strong wins for one role or another (mostly with correlated ratings already, or with no significant role-correlation because they are just complex enough to hide the strong win from the current player population - this must be true for games like the breakthrough family since they do not permit draws). It's an interesting observation that statistical measurement by Tiltyard cannot (cannot possibly in fact) uncover the correct correlations without a sufficiently string (at the game in question) population of players.
The game that brought this home to me was 'Sheep and Wolf'. This is a strong win for the sheep, but it's also fairly easy for the sheep to make a critical error and lose. The net result is that currently the sheep are considered FAR the harder role on Tiltyard, yet a sufficiently competent player (several Tiltyard players are now) will essentially always win as the sheep. The equilibrium values with the player population seen up to now are strongly ANTI-correlated with the values it will converge to in the relatively near future (assuming players continue to generally improve). Right now the player population is probably very close to the tipping point for this game.
It's fairly probably that the larger connect 4 variants exhibit the same anti-correlation (to a much lesser degree) - at least if those board sizes are actually black wins. Obviously many of the games are strong wins for one role or another (mostly with correlated ratings already, or with no significant role-correlation because they are just complex enough to hide the strong win from the current player population - this must be true for games like the breakthrough family since they do not permit draws). It's an interesting observation that statistical measurement by Tiltyard cannot (cannot possibly in fact) uncover the correct correlations without a sufficiently string (at the game in question) population of players.