Author: Dieter Buerssner
Date: 19:19:13 01/19/01
Go up one level in this thread
On January 19, 2001 at 14:32:10, Uri Blass wrote: >I think that it suggests that a better strategy in games may be to decide about >the strategy based on the root position. > >I think that probing throughout the whole search may give better results when >the number of tablebase hits is small so you can decide to change this strategy >only after you see a lot of tablebases hits. I actually played allready with this idea. However, I could not detect a pattern, that suggests in which positions aggressive use of TBs is better, and in which positions depending more on search and the eval of the engine is better. As I noted, nodes per second can go down quite dramatically with many TB accesses (Because of my hardware restrictions, I cannot test with big hash or with big EGTB cache). Nevertheless, sometimes the slower performance yields in better results, and sometimes in worse results. Of course, the the nodes per seconds is no good measure for the performance here. Better may be the time, at which a certain depth is reached. This time can also very much vary on the TB probing strategy. Also, in general, reaching the same search depth with "aggressive" TB probing should yield in more reliable analysis. So even the search depth is not a good measure. BTW. Selecting the "TB-probing strategy" dependent on the position could be considered as some sort of preprocessing. I do no preprocessing for eval at all in Yace, but some very restricted preprocessing for the search (the moment, at which some search extensions are executed can depend on the initial position). This will yield in hash table inconsitencies. I believe, that these inconsitencies will not hurt much, but I prefer to avoid them. Selecting the TB-probing strategy based on the root position will add yet another inconsitency here, when the engine does not clear hash tables between moves. -- Dieter
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.