Author: Peter Fendrich
Date: 08:51:44 08/19/04
Go up one level in this thread
On August 19, 2004 at 11:01:14, Jonas Bylund wrote: >>Uri's point is absolutely valid. The more sofisticated knowledge you add to your >>engine the harder it is to test. When you, above that, add the condition "1 >>month games" it is practically impossible to verify if the new piece of code >>it's ok. Numerous times I have added things to my engine, convinced that it was >>an improvement, just to find out that it did worse. It could be due to a bug or >>a mental black out that made me misunderstand how the code is changing the >>search tree. Whatever it is I have to catch it with sometimes extensive testing. >>I know that all engine programmers have done the same, maybe that I do it far >>too often however... :-) >> >>/Peter > >Isn't there in "normal" engines condensed code in terms of knowledge that you >would like to add but the cost in kn/s is too high, knowledge that you know is >sound, but it will not work for your engine in practical play at popular time >controls? Absolutely. There is "chess knowledge" (as opposed to knowledge about search and similar things) that I keep out or simplifies becuase of the too high cost of depth in search. It's not clear to me that the cost in general would be relatively less in longer time controls. Sometimes it is and sometimes not. The kn/s will still be low even after thinking a long time and the cost is still there. Will the cost be relatively smaller then? I don't know - we can argue for months about this without knowing... >My suggestion was not to cramp in all the knowledge at once, but to gradually >add the knowledge to the extend where you might even see an increase of strength >in 40/2 20/1 30 min for the rest games, but a decrease in strength in blitz and >rapid games. The whole point was if it was possible, if you don't care about >losing strength at shorter time controls, to increase strength at _long_ time >controls/analysis by gradually adding knowledge. In theory for some part of the "chess knowledge" yes, but in practice hard to verify and in beforehand to know exactly what kind of knowledge we are talking about. When going from blitz to 40/2h it's doably without problems but when we talk about very long time controls, like days and weeks we'll get the problem with verifying. /Peter
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.