Author: Brian Richardson
Date: 12:16:19 02/25/00
Go up one level in this thread
On February 25, 2000 at 14:17:24, John Coffey wrote: >Is there any value to storing the result of each position that we look at so >that if play that position in another game then we could resume the analysis >from where we left off last time? (I.e. but the positions in a big table. One >game of 50 moves would store 50 positions.) > >The problem I see with this is that it bypasses interative-deeping. >iterative-deepening is useful for building a transposition-table to get near >perfect move ordering on deeper searches. > >John Yes, but not EVERY position--check out Crafty's learn.c source. Many programs use learning to fine-tune opening books and also to try to avoid bad lines earlier. It should not have to bypass iterative deepinging, it just pre-fills the hash transposition table earlier so searches are faster. It can also be used to "plug" the score for a position to make sure it is or is not played. The real problem is how many positions to store (usually when the score changes significantly), and for how many moves out of book, and how to efficiently search during regular searching without incurring disk overhead.
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.