Author: Bruce Moreland
Date: 17:08:02 03/08/00
Go up one level in this thread
On March 08, 2000 at 15:04:32, Robert Hyatt wrote: >On March 08, 2000 at 12:12:07, Bruce Moreland wrote: >>I've been explaining for years how to fix this. Any 2x rep in the search is >>counted as a draw, and any 2x rep in the game history, plus one rep in the >>search, is also a draw. The only case that is now considered a draw, which >>would not be considered a draw under this system, is one rep in search plus one >>rep in game history. That's no longer a draw. > >that is one way to 'fix' it. but the fix is not exactly free. this makes >it harder to see some draws, hard enough that it is possible to not see them >until committed. What ends up happening is that in the general case, you allow a few nodes to be scored via search rather than scored as draws. I don't think that anything gets significantly harder to see. There are a few nodes that you can't immediately cut off on anymore, but those are what cause the bug in the first place. If you search those nodes, you might get a score that is below alpha, in which case you'll avoid them anyway. If you get a score that is above alpha, you'll steer for them, but in that case you'll arrive back here with two repetitions in the move list, so you won't allow these positions a third time. I have a hard time understanding why this is much of a problem. >I don't mind letting someone do this trick, as I haven't seen a case where it >has cost me anything at all, yet. And the added speed of just saying '2x is >a draw' lets me find draws quicker... I haven't tried to figure out if it's much speed you lose by doing this, but it shouldn't be that much, should it? You add one "legal" move two plies in to the search. And I have to believe that with the bug left in, it could really cause problems. As I've said before, I know of one confirmed case. I also know of another case that I suspect ripped me. The killing case is the case where you are down significantly, but find a reversible move that allows you to make a shot that brings the score up, but still below the contempt score. A savvy human could figure out that they have perfect protection against these kinds of blunders. bruce >>The benefit is that you never have to worry about this very stupid problem >>again. >> >>The drawbacks are: >> >>1) You may not allow someone who is beating you to take a repetition. >> >>2) You may play into repetitions yourself, which annoys your opponent and in >>some cases may cause you to draw games via the 50-move rule. This is actually >>fairly serious, and I still use the old method for very low material endgames >>such as KBN vs K, if I'm not using tables. >> >>bruce
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.