Author: Don Dailey
Date: 11:32:26 02/08/98
Go up one level in this thread
Hi Stuart, This is still a mystery. Langs search seems different but surely he uses some of the same principles. I remember him once saying (years ago) that no one tries to be selectivity. He described his search as throwing out 10% 20% 30% etc. moves progressively at each level. He indicated that this was the principle, not the actual algorithm. I asked him what he did if a piece was lost after looking at only 10% of the moves and he said he simply looked at all the moves in this case! Duh! But this leaves out quite a bit. Like how can you find attacking moves? If I wrote a program that statically evaluated every single move, sorted them and then tapered them something like this, it would play horrible. But Bob once had a selective program that was not based on null move prunning. He will probably tell you it's no good but at least you might get an idea of what was tried several years ago. The Chess challenger 7 had a highly selective algorithm. It played pretty lousy but was not too bad for it's day and hardware. It had a non-existant endgame though. If you were a couple pieces down you still had a good chance of winning if you could get to an endgame! The documentation that came with it descibed how many moves were discarded at the various levels. But of course it said nothing about how these were chosen. I would be fun to write a truly selective program. I'll define selective as meaning no window considerations whatsoever. Some routine just looks at the moves and keeps a few threats and the positional moves that seem ok and throws out the rest! I wouldn't know how to do this well. - Don >Hi Don, > >I don't know about selective search. I recall years ago John Stanback >talking to me about how he had wanted to do what the commercial guys >were doing by adding selectivity to his program. He spoke of this for >about a year and came back with a conclusion it was worth 100 points >to have fully implemented and working correctly and that he had done >it. At the time, I didn't have enough sense to ask more. John doesn't >write in the newsgroups much so I'm left to ask about it of others. > >I use extensions (check evasion, recapture, threat to promote -- and >have tried but discarded many others), null move (with a lot of special >circumstances where it is not used), and the other things, but not >traditional Lang/Mephisto-style selectivity by limiting the number of >moves after a number of full width ply, e.g. search 3 ply deep full >width then 8 ply selective. My question is how are those selective >ply managed -- how do they throw out the moves? I can't believe it is >knowledge based. This is the same thing MacHack did, searching wider >at the less shallow plies and then more narrowly at the deeper plies, >which was pretty much debunked in the Chess 4.x days that followed. > >--Stuart > >On February 07, 1998 at 12:48:44, Don Dailey wrote: > >>Hi Stuart, >> >>Your post confuses me. You asked about selective searches and then I >>find out you know all about it. But then you want info on Langs search >>and we have just posted all kinds of stuff on this. The bottom line is >>that we do not know, we just have some specualtions. Re-read the posts >>of the last month or so. >> >>If you are looking for some secret techique everyone but you knows >>about, then I'm not aware of any. Everyone has their own little bag of >>tricks they swear by and any single one is probably not worth much by >>itself. To write a strong program you must do a lot of things slightly >>better than the others! When I figure out how I'll let you know! >> >>I'm always looking for the mythical 1 line change that add 100 rating >>points to the program but haven't found it yet (except when it's taking >>out the latest bug!) >> >>- Don >> >>P.S. It is unclear whether Langs search "secrets" are any better than >>good null move pruning. His program is impressive (and different) but >>seems to be in the same league as the other top 4 or 5, most of which >>use null move prunning. >> >> >> >>On February 06, 1998 at 20:23:44, Stuart Cracraft wrote: >> >>>On February 06, 1998 at 18:44:35, Don Dailey wrote: >>> >>>>On February 06, 1998 at 14:02:11, Stuart Cracraft wrote: >>>> >>>>>A number of years ago, commercial programs like Lang's, and others >>>>>seemed to get about 100 points stronger due to being selective >>>>>searchers over their brute-force full-width, with capture quiescence >>>>>counterparts. >>>>> >>>>>Is this true? If so, what are the nature of the changes involved? >>>>> >>>>>Thanks, >>>>>Stuart >>>> >>>>Hi Stuart, >>>> >>>>You need to know about null move searching if you are writting a chess >>>>program. There are some good articles in ICCA past issues and you should >>>>definitely try to dig them up. Just about every program now uses it but >>>>there are a few exceptions. No one knows for sure what Richard Lang >>>>does but it's likely to involve similar ideas. If you want I can send >>>>you the basic idea and some psuedo code. >>>> >>>>- Don >>> >>>I do use null move currently and wouldn't be without it. Have been >>>using it for quite a while. No problem with it. Quite a few exceptions >>>for when it is not used though and I think I have a number of them. >>> >>>Sure, send anything you might be illustrative. >>> >>>--Stuart
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.