Author: David Franklin
Date: 06:08:59 05/04/00
Go up one level in this thread
On May 04, 2000 at 08:13:51, Steve Coladonato wrote: >On May 03, 2000 at 18:26:09, Robert Hyatt wrote: > >>Alpha/beta finds the best move and only proves that the other moves are >>worse, without proving how much worse they are. To do this requires a lot more >>time. > >You have both given me essentially the same answer. I've never looked at the >code for a chess engine so I don't know exactly what Alpha/Beta does. But the >answers here are confusing to me. I was under the impression that the best move >was determined by calculating the eval for the candidate moves. Your answers >are implying that that is incorrect and something else is used to determine the >best move, not the eval for the position. But if that is the case, is not what >the program calculates somehow related to the eval? And if so, saving the >result in an array would not incur that much more overhead so that the program >would know what the top three moves are or rather the order of all candidate >moves based on whatever it is calculating. > The point is that if all you care about is the best move, you often don't need to evaluate the other moves that carefully. Here's a hypothetical example: I evaluate move A as +10, and start analysing move B. Very quickly I find my opponent can make a move in reply to move B so that move B is worth only +9. At this point, I *stop* looking at move B, as it can't possibly be as good as move A. But I don't *know* the value of move B at all, just that it's <= +9. There might be another response to move B that is devastating, so that the true value is -999, but I don't spend time finding out. So at the end of the search, all I know is the score for the best move. [Note that these "skip the full search" decisions happen all the way down the search tree - the overall speedup is *huge*; it's not something you want to avoid doing].
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.