Author: David Rasmussen
Date: 10:00:19 01/16/03
Go up one level in this thread
On January 16, 2003 at 12:40:47, Bruce Moreland wrote: > >It can't be done with a standard test suite, because it's a test "situation", >not a test "position". > I was not asking for a test suite. >The example with the knight fork is perfect. I described something similar >without providing a diagram. > I understood your example, but I still think it is possible to design a position that will demonstrate the problem, if one does the right thing. What happens in what you describe, is that we mix knowledge that is supposed to be "perfect" (that is, search knowledge, we assumbe that both players will always make the "best" move), and imperfect knowledge, the positions on the board. If the opponent has made an error, he didn't choose the best move from the position, so the logic breaks down. The position in which the error happened, is regarded as a draw if encountered again, because what it is assumed that what was moved the last time, was the best. This assumption doesn't hold when the opponent has made an error. >The problem with ignoring positions that are "one in the tree, one in the >history", is that the program tends to diddle around, because it can make 2x >reps without suffering a penalty. > >I didn't provide any examples of a program diddling around. > Provided with a position such as the one I described above, one could enter some moves into the engine, so these moves were "over the board". One of these moves should be an error. Then one could enter some more moves until reaching a position where the program doesn't feel too good, but where it can choose to go back to the position in which the erroneous move was entered. With one kind of implementation, the program will choose to go back to that position because it thinks it is a draw. With another implementation, it would choose something else because it realizes that it isn't a draw yet, and that there is a much better move in that position for the opponent, that would make the position even worse for the program. /David
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.