Tuesday, July 29, 2014

Tired but happy with the first day

The first day of the 2014 competition consisted mostly of puzzles and non-fixed-sum 2 player games (with a couple of fixed sum ones thrown in to leaven the mix).

These were a mixture of factorizable and non-factorizable games, with a few latch-optimizable examples thrown in for good measure.

The day was hard-fought, and we started off slowly, with most of our worst games happening early on, notably a game of 9-board TicTacToe as second player (it has a fairly strong first player role bias), which was our only complete loss of the day (this and Pentago were both played with roles reversed to remove the bias).

Things then started playing to our strengths.  Most significantly a Sudoku to solve, which leveraged some significant latch and dependency analysis we perform (which will be the subject of a future post), and which only we and one other player were able to solve.  By the time we came to a match of Hex, our direct competitor on the leader-board was a player called 'General' (who's author has promised to post some descriptive information after the tournament, which I'll edit a link in to when I get it), who was also our opponent at Hex.  Fortunately for us, we were able to generate a much better sampling rate to feed our MonteCarlo tree search (about 70K simulated games per turn, against only a couple of thousand), and that enabled us to win reasonably easily.  I should note that 70K playouts per 20 second turn would ordinarily be considered a pretty poor sampling rate, but Hex is notorious for being hard to simulate quickly.

At the end of the day our we scored a total of 1612 points over 19 games (from a possible total of 1887 given the games concerned, as one puzzle had a max possible score of 87), putting us comfortably in the top spot.  This means that tomorrow we'll have the benefit of playing in the top bracket of the double-elimination phase, which means it takes two match losses to eliminate us (which is a significant advantage for the top 4 players, seeded to that bracket.

1 comment:

  1. I note that 1787 is actually the maximum possible score, because one of the 19 games that we played had broken rules meaning that nobody could score anything other than 0. Overall then, we scored 90% of the maximum possible.