Author: Robert Hyatt
Date: 07:00:03 06/22/02
Go up one level in this thread
On June 22, 2002 at 08:22:10, Filip Tvrzsky wrote: >On June 21, 2002 at 23:23:41, Robert Hyatt wrote: > >>On June 21, 2002 at 23:05:25, Filip Tvrzsky wrote: >> >>>Let me suppose my solution of your problems with hashed matescores (quite simple >>>and in fact identical with that of Mr. Hyatt): adjust the evaluation every time >>>just after you become it from the depth "depthroot + 1". >>>Something like: >>> int Evaluation = -Alfabeta(-Beta, -Alfa, Horisont_alfabeta - 1); >>> if (Evaluation < -MATVALUE + 100) Evaluation++; >>> else if (Evaluation > MATVALUE - 100) Evaluation--; >>>Of course, when the side to move is mated, you should return evaluation >>>"-MATVALUE". >>>And that is all! Then you do not need care for it more when hashing or using >>>hash values. I use this solution in my chess program now (I had also big >>>problems this kind for a long time) and it works fine! >>>Sorry for my poor English. >>>Filip Tvrzsky >> >> >>That looks problematic. alpha/beta values will be wrong, if I understand >>what you are saying correctly. That ought to make the search wildly unstable >>in a position with _lots_ of mates in various numbers of moves. > >Of course, you are right ... >I have forgotten to add the second part of this solution - I adjust Alfa/Beta >values immediately after enter in Alfabeta function: incrementing them when they >are greater then MATVALUE - 100 and, vice versa, decrementing when < -MATVALUE + >100. >I hope now it is really all. I am sorry, but it was 5.00 a. m. local time, when >I was writing my message ... The only complaint I would have with that is that you are doing a fair amount of "adjustment" at each node. Including the "if it is a mate score?" test to see if you need to do anything or not... It seems to be more efficient to do the search normally, and only do the test/ adjust stuff when actually storing an entry or getting a hit on a probe...
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.