Author: Paul
Date: 13:02:10 02/03/01
Go up one level in this thread
On February 03, 2001 at 12:58:22, leonid wrote: >I am not sure if I can respond correctly to your question. For now I have some >almost fixed four versions of selective search. Usually I use only default 4 >levels. One of them find in 11 moves and quickly. But when I go to 10 moves my >program can find the mate only with last and longest selective search. My program is the other way around I think. Simply put, it starts searching with a depth of 1 ply, then 2, then 3, etc. So the longer it searches, the deeper it gets. You can find a lot of pages on the web describing this kind of search, it's called iterative alpha-beta search. I put in a number of tricks to speed up this search even more, a lot of them you can also find on the web. And besides that I can change all kinds of parameters before starting the search to tell it what lines to extend, and what lines to shorten. For example: look a bit deeper into lines with checks in it, or with only one/two/three possible replies. There are all kinds of possibilities, but I won't tell them here, because Dann might be listening ;) Shhhhhh ... >For hash I can say nothing. I still never installed it. I wait for my next >computer to do this and probably I became too lazy as well. What I can say you >for sure is that very often search by "lighter" selective search in bigger >number of moves give better (shorter) time. In few occasions it is not so. Hash should help you enormously, at least it did me when I programmed it in. Especially in the endgame, with only a few pieces left on the board, Pretz now gets very deep! But I remember that you like boards full of pieces, so maybe it won't help you :) No, just kidding of course ... hash tables are great!! >What kind of different search you have for mate? Only if this is not the state >secret. Well, as I said above, I don't really have a different search for mate. It's part of the general program, but what I do before trying to solve a mate, is try to choose the right program settings for that particular mate. That's not always easy! And some mates (in 16/20/46) of the last weeks, I cannot solve, they simply would take too long, so I'm experimenting at the moment with a few other things. >In mine (I went through huge number of possible selective search >versions) and left for now in program (part that search for mate) very simple >variation of selective search. All difference between all those version is that >they make complet search for different number of plys before going into real >selective search. Before I had, for instance, one version that was seeing if >certain move affect enemy "vital lines" when making its search. Beyond selective search I have one brute force as well. Ok ... I understand more of your program now, thanks. I think our programs are pretty different! >No reason to be ashamed. Mine could not see in other position when your found in >few moves sooner. Difference between all possible variations of mate searchers >is that one will be good in some situations when the next in others. What >version of selective search is the most valuable could be found only through >long statistics. Not at all by solving fantasy positions like mine. Similar >positions are very good for fun but even more for cleaning all kind of bugs. >After what you write, I can see that your is rock solid. Many programs (could be >as far as 50%) can suffer from different allergy reaction and even prematurely >die in similar situations. Thanks for the nice words. But some of your selective searches are really fast! I can't really challenge that with my general chess program. And I don't intend to build a real mate solver/prover like you & Heiner have done, because I mostly use it to replay and analyze chessgames between humans, and try to find the combinations & mates they play (and miss :). Greetings, Paul
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.