Author: Lance Perkins
Date: 10:18:19 10/05/05
Go up one level in this thread
On October 05, 2005 at 04:38:05, Joachim Rang wrote: >On October 04, 2005 at 20:02:37, Jon Dart wrote: > >>On October 04, 2005 at 19:23:49, Joachim Rang wrote: >> >> >>>what has this to do with book-learning? >> >>I have two kinds of book learning. One is opening >>learning based on scores during the game. Then there >>is a separate learning pass based on game result. >>They are independent. The first strategy avoids >>lines that give a bad score soon out of book. But >>sometimes these are not actually bad openings. The >>second strategy avoids lines that lead to lost games. >> >>--Jon > >I don't get it. If your score is say -10 and the opponent looses on time you >decide that was a good opening? If the game is over just take the last eval and >decide whether it is a loss, draw or win. Then you make your book-learning. I >may even suggest margins: score >300 = win; 300 > score < -300 = draw; score < >-300 = loss. > >Joachim The point is: guessing the result of a game is not a substitute for the real result. The loss due to time is just an example. How about a resignation by a human oponent, or an agreed draw between the human oponent and the engine operator, etc, etc. Here is one example: One side is up by a Queen, but the position is locked, such that no side can make progress, so both the human player and the operator agree to a draw. In your solution, you want the engine to record this as a win for the side that is up by a Queen. A flawed solution, obviously.
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.