Previous: Automated Efficiency Optimisation
Up: Constructing a Metagame-player
Next: Advisors
Previous Page: Automated Efficiency Optimisation
Next Page: Advisors

Meta-Level Evaluation Function

With the search engine in place, using the optimised primitive operations, we have a program which can search as deeply as resources permit, in any position in any game in this class. The remaining task is to develop an evaluation function which will be useful across many known and unknown games.

Following the approach used in HOYLE [\protect\citenameEpstein, 1989b], we view each feature as an advisor, which encapsulates a piece of advice about why some aspect of a position may be favourable or unfavourable to one of the players. But as the class of games to be played is different from that played by Epstein's HOYLE, we had to construct our advisors mostly from scratch.

In terms of the representation of the advisors, we follow an approach similar to that used in Zenith [\protect\citenameFawcett and Utgoff, 1992], in which each advisor is defined by a non-deterministic rule for assigning additional value to a position. The total contribution (value) of the advisor is the sum of the values for each solution of the rule. This method of representation is extremely general and flexible, and facilitates the entry and modification of knowledge sources. We have also tried when possible to derive advisors manually following the transformations used by Zenith [\protect\citenameFawcett and Utgoff, 1992] and CINDI [\protect\citenameCallan and Utgoff, 1991], two systems designed for automatic feature generation.