Author: Dieter Buerssner
Date: 13:49:00 11/07/01
Go up one level in this thread
On November 06, 2001 at 11:54:24, Robert Hyatt wrote: >On November 06, 2001 at 11:10:43, Dieter Buerssner wrote: > >>On November 06, 2001 at 10:21:04, Robert Hyatt wrote: >> >>>If you find that the present position (at the root) is a mate in N, but after >>>you try each legal move you can't find a TB score of mate in N-1, then you >>>_must_ have a missing TB. Just turn the tablebases off _completely_ as the >>>promotion should be findable by normal search. >> >>You don't need to turn them off completely. It is enough, to not probe the >>number of pieces currently on the board. So, in a 5-men situation, probing 4-men >>tables will not yield in any problems. >Actually it can. The problem is that your search might think you are not >doing so well, and take a 4-man draw, rather than a 5-man position that looks >bad but is actually winning. I cannot see this. As in the above context, one would only disable 5-men TBs when the root position is won and a shorter mate is missing from the TBs. I think, we are both very well aware of the problem. Still I think, using 4-men TBs in this situation can only help. Actually, in all positions I tested, it did help. For example in the position, that started this thread, a mate is found almost immediately for the promotion move, with the help of 4-men TBs. Of course, the right move would be found without the access to 4-men TBs as well, but probing 4-men (and just disabling 5-men TBs) cannot hurt. I cannot imagine, how the search could see, that you are not doing well with 4-men TBs, while it will see, that you are doing well without 4-men TBs and find the correct move. If a 5-men situation (where the only correct move is a promotion, for which no TB is available) should exist, where the search cannot find the correct solution - tough luck. I have not seen such a situation, but not supporting the search with 4-men TBs should make such a situation more likely rather than less likely. >I think if you do without some, you have to do without all, otherwise you get >mixed up evaluations. How? I do all my testing with inclomplete TBs (KPPKP, KRPKR and KPPKR being the ones with pawns, but no 5-men tables with Qs at all). After adding TB-code the problem with the missing promotion table was found soon. After disabling 5-men TBs, when the shorter mate is missing (and also taking care of 5-men TB positions stored in the HT), I have not seen any problem anymore. BTW. I think the problem of some commercial engines is increased by the GUIs, that handle the situation for the engine. Also, at least some engines seem to stop very soon, when they see the first mate. Searching for the "normal" time, in most cases a shorter mate would be found anyway. When not, repetition logic should force the correct move sooner or later in most cases. Afterall, there is no problem with the 50 moves rule, because we have a promotion move available. And, the inferior (non promotion move) will still yield in a won position. Regards, 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.