Author: Robert Hyatt
Date: 12:25:00 06/10/98
Go up one level in this thread
On June 10, 1998 at 10:44:05, blass uri wrote: > >On June 10, 1998 at 06:39:35, Amir Ban wrote: > >>I don't know anything about suicide chess, but at some level it becomes >>poor strategy to assume that your opponent will make mistakes. > >I think if the computer see that the opponent has an adventage of more >than >2 pawns it is better to assume the opponent will make mistakes >otherwise the computer can do moves that do not give it a practical >chance I think this will get you killed. IE what if your opponent is simply out-searching you, and reaches a position where it loses material no matter how deeply it searches. If you assume it will make mistakes, it will hand you your head in a paper sack. > >there are some assumption that can be done > >1)you can assume in this case that the opponent must do in the next move >the move the computer would do if it had only 1 second per move. >2)you can assume the human opponent will miss long moves. > >but I think it is not very important for top programs because >in most of the cases these ideas will not help against strong opponents. > I think this is too dangerous. And it will cause you to lose games against deeper-searchers, simply because of the luck factor that will occasionally get you into a winning position. But when you incorrectly deduce that this is not luck on your part, but, rather, is a lack of skill on your opponent's part, look out... >> It's ok >>to try to create the sort of game where you are more comfortable than >>your opponent, but in reality strong humans are very good at doing this, >>while computers, despite all intentions of the programmers, usually >>cooperate. >> >>When your opponent is in time pressure and you are not, you want to make >>things complicated, but to play quickly yourself is an elementary error. > >I think for computers it is not a mistake to play more quickly than >usual >because computers are better than humans at blitz >I think that if the opponent is in time trouble the computer should >use more time than the opponent but less than the usual time >(I think a geometric mean of the time the computer has for a move >and the time the human has for a move is a good idea of the amount of >time the computer should waste). > >another thing that should be done if the opponent is in time trouble >is to increase the value of the computer's pieces in the evaluation >function >so the computer will be against trading pieces and against a draw by >3 time repetition. >I think increasing the value by 10% if the opponent has 1 minutes for >all >the game is good. >the number of % of change in the value of pieces >can be proportional to 1/x when x is the number of >minutes the opponent has for all the game. > >Uri nothing wrong with tuning your time allocation, or your evaluation (look at DrawScore() in crafty, if not playing a computer, I dynamically adjust the "contempt factor" on the fly to reflect the stage of the game, how the clocks compare, and so forth... I hate to draw humans, and am willing to accept some positional weaknesses to avoid doing so. But against a computer, that is a recipe for losing. Several "human operators" found that they could set their contempt factor to something like +.5 pawns, and use that to beat Crafty on ICC, using their favorite program, because Crafty would accept up to -1/2 pawn in positional damage to avoid drawing. I modified this to consider whether the opponent is silicon-based or carbon-based and haven't had problems in that regard since...
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.