Author: Dann Corbit
Date: 12:38:12 07/11/00
Go up one level in this thread
On July 11, 2000 at 15:19:29, Robert Hyatt wrote: [snip] >Minimaxing simply doesn't work. I did this in Cray Blitz, and in early Crafty >books. It simply allows scores to be backed up way too far... with no regard >for other moves that are not included. You end up playing down lines that are >going to get you murdered. Because one key move is not present and so didn't >influence the minimax score. > >A cap score used "at the point the position is reached" is perfectly reasonable, >as all that is missed is whatever the program producing the analysis missed. >But when you start minimaxing, you make the errors multiply... You obviously can't minimax because the majority of the data will be missing, but you can still improve the score by narrowing the window or invalidating it. Suppose that some position has an evaluation of -100 and a suggested move of Rg3. You might look at the resultant position after Rg3 and notice that the score has now become +32767 which means you are toast. You then look at all the other calculated alternatives (or ignore all the data and calculate like a madman) or some such. Let's suppose that you have some other position analyzed to 13 plies. It suggests b4, but you have 12 other possible target positions analyzed. You look at each of the one ply forward positions and see that the worst score for the opponent (which must be 12 ply or deeper to have any validity) is with a different position than the b4 choice. So you make that move instead. I think this kind of approach can be used to look for tactical snares or benefits. However, I would not replace the original data, but (rather) have an extrapolated pv and ce which would possibly be useful as an additional criteria for making a choice. Perhaps we could call them epv and ece.
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.