Author: Uri Blass
Date: 07:01:25 09/11/01
Go up one level in this thread
On September 11, 2001 at 07:58:06, Gian-Carlo Pascutto wrote: >On September 11, 2001 at 02:26:58, Uri Blass wrote: > >>On September 10, 2001 at 21:15:38, Dave Gomboc wrote: >> >>>On September 10, 2001 at 05:31:18, Uri Blass wrote: >>> >>>>No >>>>If you extend a full ply for every singular move by 0.25 pawn you will never >>>>finish your search. >>> >>>Sorry, Uri, but that's simply not the case. >> >>Deep thought needs to refute more than one line in order to get a score of +2 >>and if the singularity in most of the nodes is only singularity of 0.25 there is >>no singularity in order to prove the +2 score. > >As I have stated before, once the singular move drops in score, for example to >-2, then the others are reevaluated for singularity. As I exoplained the number of lines is too big and it makes it impossible. > >Because of this and the hashing, there can be multiple singular moves at a node, >and there can already be known singular moves at nodes that haven't actually >been searched yet. > >-- >GCP The tree is simply too big Deep thought had to prove that 28.Rd2 28.c4 28.b4 28.Rfe2 are losing more than 2pawns and these are not the only non trivial moves at move 28. 29.Rd2 29.h3 29.c4 29.h4 There are a lot of moves that lose material but clearly less than 2 pawns at moves 30,31 30.Bf3 31.Bxf5 31.Re1 White had also alternatives at move 32 except Bg5 that had to be considered (32.h3 32.Re2) It is clear that if you count the moves that were not played in the game you will get at least thousands of lines in a tree that proves +2.xx if there is a tree that proves it. Deep thought could not search more than 10^9 nodes in that game. Singular extensions cannot explain a situation when there are thousands of lines that you need to refure in order to get the score that you want when other programs cannot see the refutation of everyone of them even after a long search. If singular extensions can see +2.xx score it means that singular extension could see with less than 1,000,000 nodes things that top programs cannot see even after hours and not in one position but in many positions in the middle game. It simply does not make sense. It is known that most of the nodes of chess programs are simply unrelevant lines to prove the score (for example analyzing moves of black that are not the best moves in order to prove singularity of black moves). Uri
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.