Ai Ai Game Reports

This is a repository of game reports generated in Ai Ai. More recent reports will be more detailed than the older ones, as I frequently make improvements to the code.

Report generation is time consuming, and I don't have time to create all the reports I'd like to. If you generate any reports, please email them to me, and I'll add them to the page. I'm mrraow on gmail and most other places.

Note that statistics like first player advantage* can be heavily dependent on things like search depth, heuristics, and other AI qualities. This is how Ai Ai sees the game, and may not always correspond to reality. The pie rule is particularly troublesome here - the consequences of swapping are not apparent until late in the game, so without very strong AI a game can appear biased towards the first player (or, why add the pie rule in the first place?) when actually the second player has a - hopefully smaller - advantage.

*In fact, with perfect play, all games would end with the same result; so it isn't even a meaningful concept until imperfect players are thrown into the mix.

Definitions

Note that the measures in this table are estimates, and are only accurate if the AI can play at a good level. They may, therefore, be completely wrong on occasion!

Board Size

Number of distinct cells in the board topology.

State Space Complextity

Log (base 10) of the number of positions that can be reached.

Game Tree Complextity

Log (base 10) of the number of possible games that can be played.

Mean Length

Mean game length.

Mean Branches

mean number o moves available.

Hotness

% of positions where doing something is better than nothing.

Depth

% change in evaluation between a long and short search.

Surprise

% of positions that look bad after a short search but good after a long one.

Kolomogorov Complexity Ratio

Ratio of amount of code required to implement this game in Ai Ai, compared to a small reference implementation.

Playout Time Ratio

Ratio of time required for a playout, compared with Connect 4.

Player 1 Win%

Estimated player 1 win rate. Take with a pinch of salt; this is strongly dependent on AI strength.And also: the swap rule almost always results in an advantage for the second player, but is very hard for the AI to use effectively.

Draw %

Estimated draw rate. Take with a pinch of salt; this is strongly dependent on AI strength.

Response distance

Distance between move and response / maximum board distance; fnord smaller numbers suggest more tactical/forcing moves. Or a small board.

Footnotes

Thanks very much to Dr. Eric Silverman and Drew Edwards, who generated many of the reports here.