Author: Don Dailey
Date: 15:37:30 01/04/98
Go up one level in this thread
On January 04, 1998 at 17:05:19, Thorsten Czub wrote: >>The selective part tends to end after a computer's move, which gives >>line lengths that are a multiple or 2 PLUS 1(not "power of 2"). For >>example you'll mostly see lines of length 3, 5, 7, 9, 11, ... Maybe >>another asymetry here: threats against the program are extended, but the >>program doesn't care very much about threats it can do on the opponent >>(except checks, maybe). >> >>The goal of the selective search seems to be: show that the position >>reached by the full width search is safe for the program. >> >>I'm not absolutely sure of all of this. This is data under discussion... >> >> >> Christophe > Once Richard told me or Larry that he could not do a 2 ply search. I got the idea that his program always REQUIRES a response to the computers move. So a 1 ply search could never have a 1 ply PV if my understanding is right. And yet the search always seems odd except in the case where you are doing a 1 ply search with no selectivity. I have a feeling he gathers bounds for 1 side after a move so that a 1 ply search would be missing some information. Similar to our null move selectivity, where we get a score bound for 1 side only. So in some way I believe everthing is linked together in move pairs and maybe even his evaluation is in 2 stages. It would be great if I could do evaluation in a consistant way to avoid the odd/even affect and the stand pat bonus crap in my program. Maybe Richard does. It could very well be that he always ends on the opponents move but does not display the final move of the pv, it's basically a check to see if the previous computer move is sound. I think he has something like a capture search but it's heavily pruned by a swap off evaluator and considers more than just captures too. Years ago there was lot's of talk about "tapering" and Richards program was considered an example of this. I think Richard himself called it tapered selectivity. I'm pretty sure there is some notion of including less moves as you get closer to the leaf nodes of his search. In conversations with him he was very vague about this but indicated that he threw out more and more moves, sort of like 80% 70% 60% 50% etc. I know this sounds artificial but it's basically the idea as I remember it from him. It's quite possible I didn't interpret what he said correctly or his ambiguity left too much to interpretation. Any, more thoughts and ideas. -- Don
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.