Author: Michel Langeveld
Date: 03:21:35 03/10/04
Go up one level in this thread
On March 10, 2004 at 06:09:55, Renze Steenhuisen wrote:
>What is a normal TT-hit-ratio?
>
>For NULL-move saving I get about: 30%
>While without NULL-move saving about: 10%
>
>Hmmm.... a bit low maybe?
In this tread you can read the hashhit ration of Nullmover.
http://www.talkchess.com/forums/1/message.html?353673
But it depends also how you measure.
I count something as a hashhit if I can
A return an exact score
B return an beta bound
C return an alpha bound
D use a move from the hashtable
So I don't count something as a hashhit if
A if the position is in the hash but does not have enough depth and because of a
alpha bound I can't use a move
B a position is an alpha bound and has enough depth but the alpha is too low and
no move is present
C a position is an beta bound and has enough depth but beta is too low and no
move is present
I am not sure if this calculating stuff is like others do it.
In other words:
scoreType probeHash(int depth, scoreType alpha, scoreType beta)
{
//reset the value of the bestmove to indicate that we did not found something
hashProbesCounter++;
hashTableType *pHashRecord = &hashTable[p.dynHash & hashMask];
do_null = true;
if (pHashRecord->key != p.dynHash)
{
hash_move[ply] = 0;
}
else
{
hash_move[ply] = pHashRecord->bestmove;
if (pHashRecord->depth >= depth)
{
switch(pHashRecord->flags)
{
case HASH_FLAGS_EXACT:
{
scoreType score = (scoreType)pHashRecord->value;
if (score >= 9800) score -= ply;
if (score <= -9800) score += ply;
hashHitCounter++;
return score;
}
//More used term is UPPER_BOUND for this
case HASH_FLAGS_ALPHA:
do_null = false;
if (pHashRecord->value <= alpha)
{
hashHitCounter++;
return pHashRecord->value;
}
break;
//More used term is LOWER_BOUND for this
case HASH_FLAGS_BETA:
if (pHashRecord->value >= beta)
{
hashHitCounter++;
return pHashRecord->value;
}
break;
}
}
}
//if we found a move only we count this as a hit.
if (hash_move[ply] != 0)
hashHitCounter++;
return HASH_UNKNOWN_VALUE;
}
This page took 0.17 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.