Compressing Chinese dark chess endgame databases

Abstract

Building endgame databases is a common practice for boosting the performance of many computer game programs. After databases are constructed, we usually apply compression to save space. In order not to decrease the performance of accessing compressed files, we used block-based compression routines such as gzip. It is usually the case that bigger databases bring more gains. The sizes of the databases are fairly large even after using state-of-the-art compression programs. We discovered that the compression ratios vary a lot when different position indexing methods are used in a raw endgame file. The intuition is that when a continuous chunk of positions has more uniform values, gzip can better compress it than that of the case of having diversified values in this chunk. We report indexing methods that can upto 79.89% in compared to a naive indexing one when both are gziped. Our heuristics can be used on other chess-like endgames.