Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: tt table progress

Author: Robert Hyatt

Date: 18:25:44 03/11/98

Go up one level in this thread


On March 11, 1998 at 19:00:13, Will Singleton wrote:

>Progress on tt table implementation -
>
>Following advice, I wrote a validate routine to test whether the
>incremental hashkey matched a value calculated from scratch.  Sure
>enough, errors popped up, which were related to ep, promotion and
>castling.  Those are fixed.
>
>I've done some early data-gathering, resulting in the following
>observations:
>
>The percentage of true transpositions appearing in the table, for which
>a value can be returned immediately as the true score of the position,
>is small.  That is, using iterative deepening, a ply 6 search can only
>find transpositions within its current search, and none from the
>previous ply 5 search.  That would hold true for scores which narrow the
>window.

actually you can get a transposition in a 5 ply search...  from the
starting position for example:  Nf3 Nf6 Ng1 Ng8 and there's a position
you should have kept from the 1 ply search you did first...



>
>The main value of tt tables is to order the search by using moves that
>have caused cutoffs in stored positions, and may do so again.
>
>As for specific observations, my node count is down to about 60% of what
>it was without tt-tables.  The number of tt hits from ply 6 to 7 (in the
>table below) increase by a factor of 3, while the overall nodes from ply
>6 to 7 increase 20 fold.  Doesn't seem right.  Amateur's node count is
>excessive, but I'm only using killers and transpositions (and some
>extensions).  Move-ordering is tt-move, killers, good captures, rest of
>moves.  But even so, compared to Hiarcs especially, the node count seems
>pretty high.

the above seems wrong.  good captures should come before the killers,
always...


>
>
>Test position results -- 7 plies
>
>r2n1rk1/p4ppp/1qb1p3/4P1B1/1p1Pp1NP/1Pp1Q1P1/P1P2P2/3RR1K1 w - -
>---------------------------------------
>Amateur -- with tt tables --
>thru ply 6 - move=Nh6 score=162 nodes=118302
>               tt hits=13204  tt cutoffs=11337
>thru ply 7 - move=Bf6 score=521 nodes=2068354
>               tt hits=32687  tt cutoffs=13807
>
>Amateur -- without tt tables --
>thru ply 6 - move=Nh6 score=162 nodes=199272
>thru ply 7 - move=Bf6 score=553 nodes=3514276
>---------------------------------------
>
>Hiarcs --
>pv Bf6 h6 Bxg7 f5 Qxh6 fxg4 Qh8+ Kf7 Qxf8+ Kg6 Qf6+ Kh7;
>score 631; nodes 120121
>---------------------------------------
>
>MacChess --
>07/33|00:00:11|   Nh6+ Kh8 Be7 Re8 Bxd8 Qb7 Nxf7+ Qxf7 Bg5;
>328862 (nodes)  tt-hits:11095  tt-cutoffs:7605



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.