|
Accession number;03A0679577
|
| Title;Improvement and Estimate of MTD-f. |
| Author;
SHIBAHARA KAZUTOMO
(Tokyo Univ. of Agric. and Technol.)
INUI NOBUO
(Tokyo Univ. of Agric. and Technol.)
KOTANI YOSHIYUKI
(Tokyo Univ. of Agric. and Technol.)
|
Journal Title;Joho Shori Gakkai Kenkyu Hokoku
|
Journal Code:Z0031B
|
ISSN:0919-6072
|
|
VOL.2003;NO.79(GI-10);PAGE.1-8(2003)
|
| Figure&Table&Reference;FIG.3, TBL.3, REF.4 |
| Pub. Country;Japan |
| Language;Japanese |
| Abstract;The best-first search is an ideal algorithm but is practically difficult to use because of the problem of strage. MTD is an advanced best-first search which explores a game-tree in a depth-first manner. MTD uses a NULL-window search at a root node in several times and is able to find a solution with less number of search node than an alpha-beta search. Previous researches reported that MTD-f, which uses an approximate minimax value at first for the NULL-window search, showed the best results. We show the nature of MTD-f for random-game trees in this paper. From this observation, we propose several methods for the improvement and analyze these performances. These method are concerned how to determine initial values of NULL-windows. As a consequence, MTD-f with MTD-step, called MTD-f-step, and MTD-f-alpha-beta can find solutions 3 percent faster than MTD-f in our experiments. (author abst.) |
|
|
|
Related Articles;
|
