Author: Rolf Tueschen
Date: 04:53:42 11/02/03
With intention I avoided the use of the term fetichism in the subject line because that could have stirred the emotions and would lead the attention in false directions. Since I am a real computerchess layman (in short newbie)I missed the discussions about the interesting observation about the deminishing good of increasing depth in plies for the decisions of the engine. In looking back I'm a little bit sad that I wasn't there because otherwise perhaps I could have influenced the outcome of the debate. Anyway, perhaps it's never too late. I will keep this short and go media in res. I quote the numbers from Rebel (coming from Ed Schroder?): "Rebel maintains a table in SEARCH.TXT. It records every change of best-move per ply. Here are the contents of one of my autoplayer PC's, it clearly shows the decreasing importance per iteration." >> SEARCH OVERVIEW >> =============== >>Depth Moves Moves Big Score >> Searched Changed Changes >> 1 83591 0 = 0.0% 0 = 0.0% >> 2 82195 32077 = 39.0% 20786 = 25.3% >> 3 81737 24348 = 29.8% 13022 = 15.9% >> 4 80428 20153 = 25.1% 7033 = 8.7% >> 5 80134 19940 = 24.9% 7969 = 9.9% >> 6 79135 17344 = 21.9% 6014 = 7.6% >> 7 78694 15903 = 20.2% 5940 = 7.5% >> 8 77710 14294 = 18.4% 4918 = 6.3% >> 9 75240 12888 = 17.1% 4769 = 6.3% >>10 67180 8933 = 13.3% 2816 = 4.2% >>11 54813 7240 = 13.2% 3223 = 5.9% >>12 35267 3679 = 10.4% 2027 = 5.7% >>13 19370 1531 = 7.9% 1363 = 7.0% >>14 9860 597 = 6.1% 721 = 7.3% >>15 5233 253 = 4.8% 428 = 8.2% >>16 3128 119 = 3.8% 228 = 7.3% >>17 2173 57 = 2.6% 151 = 6.9% >>18 1573 41 = 2.6% 83 = 5.3% >>19 1186 33 = 2.8% 64 = 5.4% >>20 982 19 = 1.9% 38 = 3.9% >>21 844 10 = 1.2% 30 = 3.6% >>22 761 5 = 0.7% 14 = 1.8% >>23 700 8 = 1.1% 9 = 1.3% >>24 640 4 = 0.6% 7 = 1.1% (End of quote) As I said I missed the debate and I am a CC lay. All I have to say when I look at the numbers above: Contrary to the implication of a "decreasing importance per iteration" I defend the following from a chess view: 1) Bearing in mind a specific limitation of depth for a particular program, a decrease in numbers is no surprise. But in chess it's NOT the stats of mere numbers that counts but the chess content of an evaluation in a particular position. 2) the numbers above are just a mirror of the actual limitation of a program and NOT representative for a sort of natural law of decreasing. One could also say that the intelligence of actual programs is not sufficiently developped yet so that the decisions in deeper depths could be given more importances. 3) From a chess view of course the programmers should guarantee that their programs had a better chance to judge the outcome (the end) of their calculations before darkness makes them totally blind. So all techniques which could increase the numbers in higher depths by definition must have a strengthening influence for the program. Summary Contrary to what the experts wanted to say with the above numbers when they concluded that increasing depths would cause a decreasing importance for the engines evaluations, I hold up the chess truth that more depth will always be advantageous. So the naivety is dangerous if we accept the apparent decrease as factual truth when in reality it's just a mirroring of the actual limitations of chess programs. Rolf
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.