Author: José Carlos
Date: 09:40:16 07/19/02
Go up one level in this thread
On July 19, 2002 at 12:13:34, Uri Blass wrote: >On July 19, 2002 at 11:52:54, José Carlos wrote: > >>On July 19, 2002 at 11:14:13, Uri Blass wrote: >> >>>On July 19, 2002 at 10:31:09, stuart taylor wrote: >>> >>>>On July 19, 2002 at 10:24:36, stuart taylor wrote: >>>> >>>>>On July 19, 2002 at 09:54:08, Uri Blass wrote: >>>>> >>>>>>world championship. >>>>>> >>>>>>I suggest that the rules can say that the games are >>>>>>played under winboard(pondering off,animation off). >>>>>> >>>>>>Every 2 programs can play 100 or even more games between >>>>>>them so the total number of games of every program can be >>>>>>at least 5000. >>>>>> >>>>>>I believe that we may get significant results >>>>>>by that idea. >>>>>> >>>>>>Uri >>>>> >>>>>Somewhat significant, I believe. Even if humans can do something decent in Blitz >>>>>timing, they would not be in any way to comparable to computers if they had only >>>>>1 second per move. But computers can reach a good few plies in that time, and >>>>>play very advanced tactics. No stupid mistakes CAN be made after a whole second. >>>>>S.Taylor >>>>Sorry, I made a mistake! One second per move, I see, was never a question. That >>>>would be great time for anyprogram to do almost anything decent. >>>> But I thought that one second per game meant much more than a second, as any >>>>move can take UP TO a second without being counted as anything at all, and ONE >>>>move is allowed one-to-two seconds. That is how things used to be, I think, if >>>>not now also. >>>>S.Taylor >>> >>>Programs that are designed correctly have no problem >>>to play 1 second per game under winboard without losing >>>on time. >>> >>>Uri >> >> What about those that clear the hash tables between moves? Wouldn't that task >>make them lose on time quickly? >> >> José C. > >They can change their program not to clear >the hash tables when there is time trouble. > >This is exactly what I do because the latest movei clear >the hash tables between moves. > >Uri Ok, that might work (not sure anyway, because not clearing the hash tables in an engine that has been designed to clear them could result in unexpected or buggy behaviour) but nevertheless, that would mean the behaviour of the program wouldn't be the same as in longer time controls, invalidating your idea to get reliable information, or at least comparable to longer time controls. On the other hand, you could suggest to use a hash table of a size corresponding to the time control, this is, very small. But then there would be a problem if the program (like I do, and I guess others too) saves some information in the hash table from one search to another. Also IID would have to be disabled (at least the way I implement it, using the hash table) and MTD(f) would be totally killed without a large enough hash table. I say all of this to support the idea that different time controls are not totally comparable. José C.
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.