Previous: Advisors
Up: Constructing a Metagame-player
Next: Weights for Advisors
Previous Page: Material Value
Next Page: Weights for Advisors

Static Analysis Tables

Most of these advisors draw on information which would be extremely slow to compute dynamically at each position to be evaluated. We overcome this problem in a similar manner as with the automated efficiency optimisation for the primitive search components, by having the program compile a set of tables after it receives the rules of each game, and then use those tables thereafter. Another advantage of this game-specialisation beyond efficiency is that after receiving the game rules, the program can take advantage of the rules to fold some of the goal tests directly into the tables.

As an example, one of the most frequently used tables is the Constrained-Transition Table. This table (used by all the static mobility advisors) records for each piece and square, the other squares the piece could move to directly on an otherwise empty board. However, it excludes all transitions from this set which can easily be shown either to be impossible or immediately losing for the player moving that piece. A transition is impossible if the starting square is one in which an arrival goal would already have been achieved when the piece first arrived at the square. A transition is immediately losing if the player would necessarily lose the game whenever his piece arrives on the destination square. While these considerations do not obviously apply to any known games, they prove highly relevant in generated games. Further details on static analysis tables are provided in [\protect\citenamePell, 1993b].