Author: Steven Edwards
Date: 10:14:07 04/07/05
Go up one level in this thread
On April 07, 2005 at 09:08:43, Stan Arts wrote: >Instead, I'm trying to combine an AB-type search with a more AI approach. >It has sort of an overhead. That will be like a conscious brain that tries to >steer in a certain direction, form "ideas" and remembers them/sticks to them. >And an AB-search part that looks at tactics, and positional consequences, like >"normal". But the overhead will still have influence on that and if a best AB- >move really steers away from the current "idea" it might look for an >alternative, and things like that. (So the search won't be very efficient, maybe >getting true scores for several moves. But that's not so important. ) This sound interesting. One possibly related idea that has been tried is to have a family of evaluation functions and a family of move ordering functions and have some analysis/recognizer select among these functions for the A/B search. >So, I hope my program will be able to do things like: >Bluff. (If I've got a good position and formed a plan, but the plan turns to be >bad later on by search, then maybe still continue the plan (depending on a few >things) , and hope the opponent won't find the refutation.) >If I've found I'll be mated by a really deep mate, not to sacrifice material to >be mated in 11 instead of 10, but to play for complication, and hope the >opponent doesn't see/find the mate. And things like that. It's not really >possible to make a regular AB-program do that. There have been some attempts at this with the use of a swindle option. (Unlikely to work against a computer opponent.) Symbolic should have one of these but it's not there -- yet. It will need one, as I have noted more than a few games between the toolkit and human opponents where the program lost a half point because it resigned in a lost endgame (due to tablebase knowledge) when the human opponent might not have found the correct continuation. >This will only have use against humans, ofcourse.. >And the search will be fairly similair to previous Neurosis's one, so I can >mostly work on this overhead thing. Hope to have some news about it in a few >months.. >Goodluck with Symbolic ! Thank you.
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.