Author: Scott Gasch
Date: 11:22:08 09/12/04
I've been thinking about extensions lately and how to restrict them in a smart way. Here's an example of what I'm talking about: Say side to move is in check and has one legal reply. Also say that one legal move is not a capture. Well, we can choose to search this subtree deeper but if we do we pretty much expect the score returned to go down. i.e. if we search the subtree to 5 ply normally and get score X back I'd expect that extending and searching the subtree to 6 ply would return something worse than X because that subtree involves running the guy's king around and maybe finding a mate. Rarely does a great attack start with a king move. So if we have a score for the current position where there's only one legal move, either a material balance score or a real eval score, then use it to decide if we should extend. I tried saying something like: 1. if we're in check with one legal move 2. and the legal move is not a capture 3. and our current score is already below alpha 4. do not extend the move You could also do this the other way around with a move with a big upside. For example, a "smart" pawn promotion (one with a + SEE score or something). If you promote but you're score is already way above beta w/o the promotion move then why extend it? If it all works out then your score will be even more above beta and you will fail high after having spent more time to determine it. I tried this with mixed results. Overall my ECM/20 score dropped. Now I am thinking about what is wrong with this idea. For one, I use PVS so doing anything based on alpha in the search could cause problems. Imagine searching some line with a=b-1, deciding to extend based on that alpha, finding something great, researching with a real a..b window, deciding not to extend based on the new alpha, and missing the "greatness". Another thing is maybe I am too confident that extending one legal move is only going to make the score lower. Maybe the check leading to one legal move (which I always extend, btw) is a delay check and one legal move helps you see past the horizon faster. Anyway, I'm thinking a lot about this stuff and would welcome any thoughts! Thanks, Scott
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.