Author: Robert Hyatt
Date: 09:05:20 01/30/06
Go up one level in this thread
BTW: The two most common horizon-effect moves are checks and captures. If I check you, you must get out of check. Even if you have my queen trapped and can capture it the next move. So if I continue checking you, until the search has reached the max allowable depth it can probe, then the queen capture was pushed beyond the search horizon, which means as far as the search is concerned, it doesn't exist and the queen has been saved. That's why we do check extensions. Yes the extensions find deeper mates, which is important. But more importantly, when I check you, I make the search go a ply deeper as well, to keep extending the search horizon beyond the limit of the checks so that the search can see that queen being taken and produce the correct score, rather than the imaginary score where the queen is "saved". This also happens for captures. If I am about to lose my queen, I might choose to play RxB threatening your queen. You therefore have to waste a ply to re-capture my rook, but now I have pushed that queen loss beyond the search horizon so that it appears to no longer exist. But it always comes back after you take my rook. You can find lots of positions from very old ACM chess tournaments where this produced funny results. In one famous case, one side won a piece (it thought) but it had to give it back. Or else give up a pawn to push the piece re-capture beyond the horizon. 3 - 1 = 2, so give up the pawn and save the piece. But next move, the same situation was there. Give up another pawn. And eventually you can't give up any more pawns, the piece is still lost, and you took an even trade and turned it into a dead loss. This is the reason for the "recapture extension". So that a cap/recap doesn't eat two whole plies and suffer from a horizon that is too close to the root. The "horizon effect" is about the program pushing things out beyond where they can be discovered by the search. And if the search can't see the thing happening, then so far as it is concerned, it will never happen. At 4-5 ply back in the 70's, this was very common. It is still there today, but the depths are so huge, the effect is not nearly as obvious, and with better search extensions (the early 70's barely did checks, much less recaptures and the like) the effect is further minimized. But the main point here is that "the horizon effect" is something the computer search actively "uses" to hide things that are unpleasant. It is not something that "just happens" because it can't search deep enough. There will always be positions that are beyond the search depth horizon of any program, but all programs have a definite search horizon limit, and they use this "limit" in ways we would never think possible or logical.
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.