Author: Robert Hyatt
Date: 07:28:29 01/17/03
Go up one level in this thread
On January 17, 2003 at 01:26:32, David Rasmussen wrote: >On January 16, 2003 at 20:03:28, Robert Hyatt wrote: > >>On January 16, 2003 at 17:01:29, David Rasmussen wrote: >> >>>On January 16, 2003 at 16:48:22, Robert Hyatt wrote: >>> >>>> >>>>See my response earlier. With 8-9 pieces on the board, you are doing a >>>>worst-case >>>>test, as many lines will turn into a probe. With more pieces, this tapers off. >>>>And, of >>>>course, if you are in a game, the cache gets "seeded" slowly so that the >>>>drop-off won't >>>>be as bad as when starting off in a near-EGTB position. >>> >>>I know this is a harsh test, but to begin with I posted this because I >>>experience this slowdown in real games. >>> >>>/David >> >> >>Everybody has seen this. That is why most of us limit how deep we are into the >>tree when probing is allowed. The information you get is perfect, but if you >>probe too much, the loss of depth gets you killed tactically. >> > >*sigh* As I've written several times in this thread, I _do_ limit probing. More >so than Crafty, for what I can read. You probe in the first two plys >unconditionally, I do it in the first only. And if you have the whole set, that >rule doesn't even make sense. All other limitations are the same as Crafty's as >far as I can see. You can also see from my probe numbers that I don't probe more >than others. I don't understand your comment. I don't "probe in the first two plies unconditionally". If I am doing a 12 ply search I probe in the first 12 plies unconditionally. But even so, it is _still_ going to slow things down. I/O is nowhere as fast as actually searching. You just get more accurate answers from the I/O to the EGTB data. But you _will_ go slower. > >>You have to find the right balance. I used to have an adaptive algorithm that >>varied the depth limit for probes based on how badly I was slowing down, but >>it was too complicated and too much trouble. And probes sometimes come in >>spurts with long "silent" spots due to hashing, etc... > >Sure. > >/David
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.