Author: Heiner Marxen
Date: 17:14:33 03/02/02
Go up one level in this thread
On March 02, 2002 at 19:43:41, leonid wrote: [snip] >Just tried this position with new selective version and, bingo! Mate in 10 moves >in zero seconds. > >Actually, before my selective version had three "mixed" searches and now I have >5 of them. Mixed search - search start with brute force and later goes by >selective. Selective - attacking side look only promissing moves and also number >of responses is limited. In my version last Summer my three version of "mixed" >search had 1, 2 and 3 moves deep brute force before selective. Now this brute >force search goes as far as 5 moves. When I asked to search in 5, "mixed" search >response was instant. I not even tried to search for 4 moves brute force search >in mixed. > >It could be that in your future selective search (if you will be interested one >day to install it) you can try something like I did. Well, sure I am interested! Unfortunately, this is more complicated as it sounds when the hash table gets involved (you do not yet have it :-), since interior nodes (positions) can be solved/asked with different kinds of jobs, which are not always clearly sorted. I'll explain... With normal brute force (or also purely selective) any position's job/result is characterized just by the depth. If a hash table entry stores "no solution in depth=5", we immediately know the answer for all smaller depths. And if a larger depth is asked for, we have to compute it, but later also will store the new result, since the old result will be implied by the new one and will not be lost. If now the job for a position is characterized by two numbers, namely the depth like before, and an additional value, how many of the moves may be unrestricted, the new job and the old job may be uncomparable. Consider, the hash entry says "no solution in 4 moves with 2 full width", and now I ask for a solution "in 5 moves with 1 move full width", then neither the old implies the new, nor the new implies the old. Which one will I store? And if I try to store more than one result, how many of them? If I have room for all of them, this costs considerable amounts of memory, which I rather would like to use for more hash entries. I have not yet found a convincing solution for this mixed results problem. Of course, there also should be some special handling in the last plies, which is doable, but a bit complicated with my current heuristics. Still, selective search is a very attractive thing, and I will do something about it, sometime. Cheers, Heiner
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.