Author: Ingo Lindam
Date: 00:31:54 12/03/02
Go up one level in this thread
To obtain rather the N-best variations instead of one single-best I think you don't have to change that much in alpha-beta. The problem is more that you have to obtain much more nodes (atleast in case your single-best-alpha-beta works well caused by a good move ordering). If you want the engine to obtain the N best variations, you just may estimate the worst of the N best scores/moves instead of estimating the best score/move in the node for that you just obtain a score. By doing this you should be able to obtain a tree of variations that contains not only the N-best variations, but the N-best moves in each node of that tree. And now YOU have to define, what you call the second best variation in that tree to prune/cut down the tree to contain only the N-best variations within this tree/this number of plys (if want to do this). You can say: "I would like to see the 2nd best variation assuming me to make 1 suboptimal decission." or "the second best variation assuming my opponent to make one suboptimal decission" or "assuming me and my opponent to make one suboptimal decission, leading into 3 instead of 2 variations." or "the second best variation caused by either me or my opponent making one suboptimal decission where the suboptimal decision is the 'smallest mistake' that could be made within the obtained N-best tree". Internette Gruesse, Ingo
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.