Author: Bob Durrett
Date: 15:05:34 12/21/03
Go up one level in this thread
On December 21, 2003 at 12:37:30, Jaime Benito de Valle Ruiz wrote: >>I don't mean to pry, but I am curious to know the criteria your software uses to >>select the PV line to be displayed. Do you somehow identify a critical path >>leading to a leaf node where the leaf node chosen is no worse than any other >>evaluated leaf node? How can you be satisfied that the PV line displayed meets >>your criteria? As a user, can I "believe" your PV? >> >>Generally, I assume that the first move in any PV is most reliable and that the >>next move is less reliable, etceteras, due to some sort of horizon effect. For >>example, the evaluation of a leaf node position must be done without the normal >>searching process. [This is true in spite of the fact that the position >>evaluation is necessarily sequential because that's the nature of >>microprocessors, ignoring multithreading.] I would expect that leaf node >>evaluations would be less reliable than the evaluations of internal nodes since >>branches are evaluated. >> >>Bob D. > > >The "criteria" is basically the same for all programs (although the pruning and >other things are different, of course): The engine reaches the current max. >depth, and chooses the "best move" for the side to move by trying and analysing >all available moves, and then it assign it a score. Then it goes down one ply, >undo the last move, tries another one, and then chooses the "best move" for the >side to play in that new position again. Repeating this process recursively, it >eventually comes up with a PV and with a score associated to it. If the analysis >at the leaves is assumed to be perfect, then the PV is also perfect. The >analysis is far from being perfect for any engine, so the PV.... >Most programs -I guess- update the PV as they search, so if there are no bugs, >you'll be able to see the real PV associated with the score; any other line will >give (according to the engine judgement) a worse score (or maybe the same, but >not better). Shredder is probably extracting the PV from the hash tables, so if >any position has been re-written during the search for some reason, you get the >wrong PV. This is just a guess. >Obviously, a bad evaluation function will lead to stupid PVs in most cases. >A material-based-only evaluation function will lead to absurd PVs most of the >time, except for when you blunder.... and then they will find the way to win you >a pawn or whatever. > >Regards, > > Jaime Thanks for the reply. I guess I now wonder whether or not other programmers have chosen to get their PV the same way you do. Thanks again. : ) Bob D.
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.