Author: Michel Langeveld
Date: 08:20:58 10/16/03
Go up one level in this thread
On October 16, 2003 at 10:51:42, Uri Blass wrote: >On October 16, 2003 at 10:27:41, José Carlos wrote: > >>On October 16, 2003 at 10:05:23, Uri Blass wrote: >> >>>On October 16, 2003 at 09:56:59, Tord Romstad wrote: >>> >>>>On October 16, 2003 at 08:35:54, Uri Blass wrote: >>>> >>>>>Nalimov tablebases are not very important for endgames and they are not going to >>>>>change results in most of the endgames. >>>> >>>>This depends to a great extent on the engine, I think. The experiments I have >>>>seen which conclude that tablebases have no measurable effect on playing >>>>strength >>>>have always been conducted with strong engines like Yace and Crafty, which >>>>probably play excellent endgames even without tablebases. I expect that a >>>>program >>>>with little or badly tuned endgame knowledge will profit much more from >>>>tablebases. >>>> >>>>Tord >>> >>>Most of the mistakes in endgames are in positions when tablebaes cannot help. >>> >>>Maybe they can help if the program does not know to win KQ vs K but better >>>evaluation can solve this problem and more problems so I do not think that >>>tablebases is the right thing to add. >>> >>>Uri >>> >>>Uri >> >> KRPKR is hard without tablebases, same for KQKR, KBBK, KBBKN, KRKN and some >>others. The importance of tablebases is not only to play those endgames >>perfectly, but to choose the correct variation many moves before when you see >>those positions in the search and your eval is not enough. > > >I know but my point is: >1)most of the endgames when weak engines blunder are decided because of mistakes >when tablebases cannot help. >2)good evaluation that can help for solving the problem in 1 can help also >to solve tablebases position so it is better to do like you do in anubis. > >> On the other hand, in Anubis I have chosen not to implement tablebases but >>good eval. IMO, a good eval is better because it's more general so you can >>evaluate endgames with many pawns, it's faster and you don't depend upon >>external files (this is important when you release your engine as freeware and >>other people don't have tablebases). >> Averno plays endgames badly without tablebases, probably worth 50 ELO or more >>(haven't checked it). >> >> José C. > >I will be surprised if it is worth 50 elo or more. >tablebases may even be counter productive without knowledge because you may >prefer KR KPPP that win for the pawns and not KR KPP that is a draw based on >tablebases. > >Uri Dieter did some months ago extensive testing with Yace about the difference of tablebases .... he could hardly measure any difference. It's more a matter of style in strong engines. Be aware the tablebases can also slow a program down a lot and ruine the searchdepth.
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.