Position-Value Representation in Opening Books
The standard position-value representation in search engines uses heuristic values, proven win, proven loss and, if applicable, proven draw. The advantage of this set of values is that propagation is efficient and that the propagated information is sufficient for selecting a move in the root position. In this paper we propose a new position-value representation which is suited for the use in opening books. In an opening book, propagation speed is not as important as in a search engine. Instead we want to have values which describe as accurately as possible the knowledge we have gathered about a position. As a solution we introduce three new value types: the at-least-draw and at-most-draw, which propagate additional information when the opening book contains positions with value draw and cycle-draw, which propagates additional information when the opening book contains cycles which lead to a position repetition.
Unable to display preview. Download preview PDF.