Author: Tord Romstad
Date: 06:47:40 04/22/04
Go up one level in this thread
On April 22, 2004 at 09:35:48, Stan Arts wrote: >>It's possible (probable) that this is an artifact of the hash table. >>Gothmog *could* be grafting on large sub-trees via the TT so that >>the apparent depth really is 41 etc. (I'm guessing like the rest of >>us you adjust the depth of all mates retrieved from the hash table). This would be a plausible explanation if announcements of "mate in at most some ridiculously high number of moves" happened very infrequently, but this is not the case. Actually, such announcements occur in the clear majority of cases when Gothmog finds a mate in 10 moves or more. >>Interesting anyway, like you, I wouldn't worry too much if it isn't >>causing you any real problems and it's actually kinda cool. >> >>--tom > >Hi Tom and Tord > >Maybe Gothmog gives those values, because/if you return beta somewhere, >maybe after a null-move. (So then it's not always fail-soft, and beta >could be that mate in 41 value.) Not a bad guess, and I think you are not too far away from the truth. But it cannot be the full explanation. In the search that returned a score of >=#41, the search bound had a value of +15.88. >The same happens sometimes in my program Neurosis, that for instance >has a current score of -9997 (will be mated in 2) and finds a better >move with a minimum window, it can rarely happen that it gives a score >first of -9996 before the research and that value is nowhere returned in >the searchtree except for nullmove and returning beta, and then opening up >my window say 2.50 it can rarely happen that it returns -9746 before >another research, and announce it's a mate in 125 or something like that, >for the same reason, but it hasn't searched a real mate in 125.. Gothmog is an MTD engine, and therefore doesn't have any windows at all, but just a single search bound which is passed down the tree. Tord
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.