Author: Robert Hyatt
Date: 07:28:49 07/10/99
Go up one level in this thread
On July 10, 1999 at 01:36:27, KarinsDad wrote: >On July 09, 1999 at 20:56:10, Robert Hyatt wrote: > >[snip] >> >> >>None of that goes with "alpha/beta minimax depth-first". I _must_ search to >>tip positions before looking at any score... And I'm not willing to bet that >>searching 30+ plies deeply along selected lines will not find some way to make >>this happen. > >Hhhmmmm. 30 plies (only 15 moves per side) would require that the pawns be >relatively far advanced already or promoted to queens already. Still, it seems >unlikely that if most of the pawns are already promoted to queens that the >program would not have found a mate 2 ply earlier in the game when it was >searching 30 ply then (or 4 ply earlier or 12 ply earlier or whatever). But, the >only way to know for sure is to record it if it occurs within a given program >(at least you would know that it hasn't yet occurred until it actually does). > You are still overlooking the point... finding a mate does _not_ mean that no other branches in that area of the tree are searched. When I try a move that leads to mate, I still try the other moves at that ply as well, as they might lead to a _shorter_ mate. And that lets me run the search way on. I have, on occasion, tested 'maxply' in crafty to see how deep it goes, and I have hit 60 plies on several occasions... 60 plies meaning that from the root (ply=0) I tried 30 full moves, 30 by white, 30 by black, which is a _long_ way from that original position. As I said, 9 queens is certainly unlikely... but I would either allow them, or make sure they can never happen by kludging the promotion code to prevent it. I would not take a chance that in some important game, I will blow up because it does happen. And since I can't prove it won't happen, I assume it can... > Remember minimax doesn't search "logical" lines. It searches >>+all+ lines. The position I reached with 5 white queens was done instantly in >>a blitz game. At the starting position both sides had _one_ queen. By the >>time I did an instant 8 ply search + extensions, etc, and got to the first >>call to Evaluate() I had 5 white queens, which produced a score way > +infinity >>and blew me up. > >Hopefully, I've already taken that into account within my evaluation (the simple >one, not the detailed one, see below), but I haven't tested it yet. > >> >>I agree that it is very unlikely. But full-width searches often act like >>"help-mate" searches... one side does everything right, the other side does >>everything wrong, leading to very bizarre positions. I'd at least cover the >>case so that it wouldn't blow up. Either by allowing 64 queens as I do now, >>or by making sure no promotion is allowed if it takes me over my max limit. > >Always better to be safe than sorry. > >I have a square control mechanism which if I remember correctly, only takes into >account up to 5 queens, 4 rooks, 4 bishops, 4 knights, and 2 pawns per side (the >kings are considered separately). Now, since I use this for square control, >obviously it is a little safer for any given square to assume this lower number >of pieces (as opposed to on the entire board), but you have now gotten me >thinking that maybe within a search, it is really not that safe at all. I have been burned too many times. Searching deeper than I thought physically possible, reaching a position with more than N legal moves, where my move list was of length N, having more than N of a single piece where N is the max due to the way I coded something. With Crafty you can fill all 64 squares with what- ever you want, and it won't cause me a bit of grief... > >This was the square control for the non-battery square attack table, so 4 rooks >max is fine, 4 bishops max is fine, and 2 pawns max is fine, but you could have >upwards of 8 queens or 8 knights. I figured that this would be tough to >accomplish in the search (especially the knights since not only would the >program have to consider promoting to a knight, but it is real hard to maneuver >a lot of knights to the proper squares so that they can all attack the same >square). But, because of your example, I may have to reconsider. Thanks Robert! > >KarinsDad :)
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.