Author: Tony Werten
Date: 00:34:19 08/10/02
Go up one level in this thread
On August 09, 2002 at 23:21:25, Robert Hyatt wrote: >On August 09, 2002 at 12:58:00, Vincent Diepeveen wrote: > >>On August 09, 2002 at 00:50:04, Robert Hyatt wrote: >> >>for 5 men, uncompressed bittables 6GB or something similar. >>Even my own bit less well compressing format it's well under 7 GB. >> >>if you compress that, it fits nearly in RAM... ...so you can probe >>anywhere. also qsearch also leaves *everywhere*. > > >Why would you want to? In the evaluation? If you make a capture, and >then don't get a hit, and make another move, why probe _again_ as you >will not get a hit there either. It is far better to probe where you >transition into the egtbs after a capture. That's only with normal egtb. If you only use 1 bit, you might get a hit that says won, but when you are searching with an open window you can't take the cutoff because you're searching for the exact mate. Tony >If you fit 'em all in RAM, >then doing this in the q-search might be perfectly reasonable. For normal >7.5 gig compressed tables, probing in the q-search is a _bad_ idea... > > > > >> >>>On August 08, 2002 at 14:43:32, Alvaro Jose Povoa Cardoso wrote: >>> >>>>If I wanted to play with EGTBs with only win/loss/draw information and no >>>>DTM/DTC information where would I place the probe code in the search routine? >>>>Taking crafty as an example it has the TB probe code placed next to the hash >>>>table probe code and if the current position is in a TB file it returns >>>>immediately because is has perfect/complete information. But with 2bit/entry >>>>EGTBs it just doesn't seem right to place the code there. It seems logic to me >>>>to place the TB probe code in the evaluation function and if a leaf node is in a >>>>2bit/entry TB file give it a bonus. Another possibility would be to place the TB >>>>probe code before calling the qsearch. >>>>Could someone tell me if I'm thinking correctly? >>>> >>> >>> >>>No. You should probe _exactly_ where I do. The only difference will be >>>the "score". You won't be able to return a mate in N, so you will have >>>to doctor the score to some value that says "mate in N where N is large and >>>unknown." >>> >>>If you probe in the eval, you will probe a million times too often. You >>>should probe when you drop into a 5 piece (or smaller) ending, which only >>>happens infrequently and immediately after a capture only... If you do it >>>at endpoints, you will get killed tactically because your search depth will >>>drop off _several_ plies... >>> >>> >>> >>>>Best regards, >>>>Alvaro Cardoso
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.