Author: Dann Corbit
Date: 15:48:05 07/28/04
Go up one level in this thread
On July 28, 2004 at 07:06:22, Uri Blass wrote: >On July 28, 2004 at 05:53:57, Tord Romstad wrote: > >>On July 28, 2004 at 00:36:12, Dann Corbit wrote: >> >>>Play with the granularity of the search. If a pawn is worth 64 centipawns you >>>will do far less searches than 1000 centipawns. But then you lose resolution. >>>So the trick is to find a happy medium, I would guess. >> >>I have actually used pawn=32 since quite a long time (if I recall correctly, >>Gothmog 0.4.6 and later use pawn=32, older versions used pawn=64), and I haven't >>seen any bad effects of the low resolution. The internal resolution in my >>evaluation function is pawn=128, but I divide all scores by 4 before I return >>them. >> >>Tord > >How much rating do you earn from dividing the evaluation by 4. > >I use pawn=100 and I do not divide my evaluation by 4 before returning them. >Maybe it is a bad idea but I am not sure. > >Note that I am not sure if tactical tests will help to find it and I guess that >the only way to know is games because it is possible that you pay the price not >in tactical positions but in positions when you choose slightly inferior >positional moves. You do not use MTD(f). If you use PVS, then a resolution of 20 significant digits is the same as 3 digits for speed. If you plan to switch to MTD(f), then you can play with that parameter. The thing about MTD(f) is that all searches are zero window. Imagine if you substitute the PV node search for a series of zero window searches instead of a full width search. If you choose 32 as your resolution, then 5 probes is the window a pawn in size in a binary search. If you chose 1024 as your resolution, then it would be 10 probes for a window 1 pawn wide. Now, a MTD(f) search is guided by the estimation, so it is usually not as bad as a binary search, since a good guess is available to start with. But sometimes, you have to do several probes after the first guess. PVS is not sensitive to resolution in the same way.
This page took 0.01 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.