Author: Uri Blass
Date: 07:02:13 01/10/03
Go up one level in this thread
On January 10, 2003 at 09:38:52, Christophe Theron wrote: >On January 10, 2003 at 05:00:39, David Rasmussen wrote: > >>On January 09, 2003 at 19:05:34, Mogens Larsen wrote: >> >>> >>>By the same argument, you could say that anyone should avoid testing engines in >>>environments where they are likely to suck. If an engine is poor at blitz, it >>>shouldn't be tested at blitz timecontrols. Likewise with any other imaginable >>>parameter. Otherwise the tester obviously suck. >>> >>>If there's a bug, it should be fixed. If the bug causes inferior performance. >>>Too bad. >>> >> >>Exactly. >> >>EOD. I don't want to go to the strange and unknown place that is the weird >>logics of Thorsten. >> >>/David > > > >That strange place is where everybody lives I think. > >Configure ANY chess engine so it uses too much hash table for example, more than >the available RAM. > >Then most chess engines are going to perform extremely poorly. Very old chess >engines that do not use hash tables on the other hand are going to look >extremely strong. > >There are other settings that you can set to extreme values and that will make a >particular chess program dumb. > >So there is an untold rule among programmer and users: we try to test the chess >programs with settings that do not handicap them. The idea is to see which >engine performs the best when they are set with optimal settings. > >You can then complain that a particular chess engine has a problem with some >particular setting, but that's another story. > >In the case exposed in this thread, I would agree that one can complain that >Chess Tiger does not work well when you use Fisher time controls. And this >should be fixed. > >But I insist that testing Chess Tiger under Fisher time controls with big >increments is like giving a big handicap to Chess Tiger. You just expose a >weakness that lives in the time management of the engine, not in the core of its >chess knowledge. So the experiment is not really interesting in my opinion. > > > > Christophe I believe that Tiger has better time management then most engines in games with no increasement so it is also possible to claim by this logic that if you use games with no increasement you give a handicap for other programs because the programmers did not work on better time management(Ruffian is one of the programs that is criticized by the ssdf for bad time management). It may be possible to use fixed time per move but even in this case authors can have better algorithms if they know that the engine get fixed time. It is possible to decide not to calculate the exact value after fail low or use different size of window in the last iteration. Uri
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.