Author: Dave Gomboc
Date: 12:52:03 12/05/99
Go up one level in this thread
On December 05, 1999 at 13:30:52, Albert Silver wrote: >On December 05, 1999 at 12:49:48, Christophe Theron wrote: > >>On November 30, 1999 at 01:32:41, Dave Gomboc wrote: >> >>>On November 29, 1999 at 09:10:26, Albert Silver wrote: >>> >>>>Guadeloupe isn't exactly a hotbed for strong chess players. I'm not sure how >>>>much he can learn from such games. >>>> >>>> Albert Silver >>> >>>I must concur, but am also compelled to comment that Christophe's war chest of >>>386s isn't exactly a hotbed of speed either. :) >>> >>>Dave >> >> >>That's exactly the point. Using a 386sx-20MHz again and again against players >>that understand where the weaknesses of the program are is a terrible test, >>believe me. >> >>For years, they wanted to play against Chess Tiger because it was fun to beat >>"the" computer. >> >>Tiger had no learning, and usually the same player tried to play the same >>unsound king attack over and over until he won. Usually it took several trials >>because the human player would do a tactical mistake. I did not prevent them >>from doing so. >> >>I have learned a lot with this. >> >> >> >> Christophe > >I understand, and this brings to mind Ed's comment sometime ago in which he >explained how hard it was nowadays to see where and when Rebel was improving or >not, simply due to the depth and strength of the program. Naturally, it is much >easier to see what it is doing wrong at 6-7 plies than 11-12, nevertheless the >difference between a 2000 rated player and an GM rated 2500 is not merely one of >depth of calculation. Naturally, the GM is calculating far deeper, but there is >more involved. Suppose your program is getting 6-7 plies in a long game on your >386 and as such you have really maximized the knowledge and performance >according to what you have available. All the same, there are probably certain >things that simply cannot be done on a 386 that could be implemented on a more >powerful computer, because that more powerful computer has an edge that goes >beyond merely doing the same thing the 386 does but faster. This is why it is >impossible to properly compare programs like DB or Cray Blitz with other PC >programs. That is why comments like "if Hiarcs ran on a Cray it would be >stronger than Cray Blitz" have no meaning, as the program is inseparable from >the hardware. I think that as hardware develops, new things are possible in >programs that weren't possible in the past, but at the same time these newer >generation programs won't be retro-compatible, because what they do is only >possible with this new hardware. > > Albert Silver This is an interesting point. Many people continue searching a line until depth >=n, where n is whatever value for that particular iteration (e.g. 1, 2, ...). I prefer to look at it as starting from n, and stopping when depth <= 0. Why? This should mean that none of your existing logic about when to extend and when to prune has to change as the speed of computers increases. It is possible to simply define the conditions for extension and pruning effects to occur at higher and higher depths as they arise, when a decent starting point is "the same as the next-highest depth". Even when you're doing 10-ply searches, all that work you put into tuning your 3-ply searches 10 years ago will still be worthwhile from more than a "general experience" standpoint, because you'll still be doing the same thing at every place in the tree where you've got three plies left to search. I would be interested in hearing from programmers whether this works or has _not_ worked for them. Dave
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.