Author: Robert Hyatt
Date: 06:29:27 01/26/00
Go up one level in this thread
On January 26, 2000 at 08:08:01, Tijs van Dam wrote:
>Hello,
>
>Looking at the crafty code, i stumbled upon something i don't understand. It is
>in search.c, in the part for internal iterative deepening. I left out the abort
>checks:
>
> tree->next_status[ply].phase=FIRST_PHASE;
> if (tree->hash_move[ply]==0 && do_null && depth>=3*INCPLY) do {
> < ... check if this is indeed a pv node ... >
> tree->current_move[ply]=0;
> value=ABSearch(tree,alpha,beta,wtm,depth-2*INCPLY,ply,DO_NULL);
OK: The normal window failed low. I am now going to do a search that I
_know_ can't fail low, because alpha=-INFINITY...
> if (value <= alpha) {
> value=ABSearch(tree,-MATE,beta,wtm,depth-2*INCPLY,ply,DO_NULL);
here the following line is missing:
tree->hash_move[ply]=tree->pv[ply-1].path[ply];
This was deleted somewhere along the way. However it isn't really
important... this case is _very_ rare.... a search to depth=N will produce
a score of X, A search to depth=N-something only rarely fails low and breaks
this...
> }
> else if (value < beta) {
> if ((int) tree->pv[ply-1].pathl >= ply)
> tree->hash_move[ply]=tree->pv[ply-1].path[ply];
> }
> else
> tree->hash_move[ply]=tree->current_move[ply];
>
> tree->last[ply]=tree->last[ply-1];
> tree->next_status[ply].phase=FIRST_PHASE;
> } while(0);
>
>I understand this: to get a good move, you do a search with depth-2. When the
>search returns something in the range (alpha,beta), you are happy and use this
>move to search first. When the search returns a value >=beta, you use the move
>that caused the cutoff, because there is a good chance that it will also cause a
>cutoff at this depth.
>
>But when you get a value<=alpha, you do a research on (-MATE,beta). The outcome
>of this search is not used, although a move must have come out of it?! Why is
>the best move from this search not used, even if it is below alpha in a
>shallower search?
>
>I want to implement iterative deepening too, but i keep puzzling over this code.
>Dear Dr. Hyatt, could you please explain? BTW, this is crafty 16.something.
>
>Greets,
>Tijs van Dam
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.