Author: Robert Hyatt
Date: 14:22:56 09/20/99
Go up one level in this thread
On September 20, 1999 at 16:57:23, KarinsDad wrote: >On September 20, 1999 at 16:30:18, Robert Hyatt wrote: > >[snip] >> >>King safety is a separate term... and don't forget, these scores are modified >>by the material remaining on the board. The less material, the more a outside >>passer or candidate outside passer is valued... And I only do the evaluation at >>the endpoints of the tree, using traditional alpha/beta, I _never_ use the eval >>to make forward pruning decisions or decisions that would say lets stop >>searching this branch here and go no further... > >So, does this evaluation get done at every endpoint? If so, with a branching >factor of 4, it would seem that you are still doing the evaluation on about 75% >of the nodes. Is this correct? > No... although I can't tell you exactly what percentage do get evaluated. But the ones that are hopeless (in material) don't get touched by the slow eval code, and there are a lot of those... IE if I am a queen down, I don't need to do a full eval as the lazy eval will bail out quickly. But in general, I "try" to evaluate at _every_ node in the q-search. But can avoid some due to the above material quick exit... >> >>This is an interesting piece of code, piece of knowledge. It will be >>interesting to see who doesn't have it now that I do, because it will become >>obvious. IE I can tell you who doesn't know about the bishop trap at a2/h2/etc, >>and who doesn't know much about outside/distant passers, because when I play >>them suddenly this term becomes important. This majority term will likely >>show up in similar ways for games that reach endgame status... > >You have a lot of GM contacts. I bet you could get from them other endgame pawn >knowledge which would be very practical. > >KarinsDad :) Some is... ie the 'hidden passed pawn' code in crafty came from a GM suggestion. I watched as he had a duo at h5/g5 and crafty had one pawn at h7. The GM couldn't really play g5 to make a passer, so he calmly played h6 and crafty thought the score got better for it because of the backward pawn the GM just created. When I asked later, he replied "I couldn't make a passed pawn on the g-file at the moment, so by playing h6, I made a passed pawn at h6 that you can't see. A protected passed pawn, in fact." When I thought about it, it made sense obviously, as white later plays g6 and gets a passed pawn no matter what black does. I'm sure there will be additional ideas... I'll try to keep them documented here in CCC as I add them...
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.