Author: Robert Hyatt
Date: 08:27:31 08/30/00
Go up one level in this thread
On August 30, 2000 at 11:04:47, Uri Blass wrote: >On August 30, 2000 at 10:40:36, Robert Hyatt wrote: > >>On August 30, 2000 at 09:45:45, Wayne Lowrance wrote: >> >>>On August 30, 2000 at 04:47:49, Uri Blass wrote: >>> >>>>On August 30, 2000 at 04:34:08, Ricardo Gibert wrote: >>>> >>>>>On August 30, 2000 at 02:42:49, Uri Blass wrote: >>>>> >>>>>>On August 30, 2000 at 00:31:24, Ricardo Gibert wrote: >>>>>> >>>>>>>On August 29, 2000 at 23:19:59, Robert Hyatt wrote: >>>>>>> >>>>>>>>On August 29, 2000 at 19:18:17, Alexander Kure wrote: >>>>>>>> >>>>>>>>>On August 29, 2000 at 13:58:52, Graham Laight wrote: >>>>>>>>> >>>>>>>>>>Firstly, apologies to everyone for dashing off after the last game in the WMCCC. >>>>>>>>>> >>>>>>>>>>It enabled me to get an extra day's holiday with my girlfriend, though, which >>>>>>>>>>was well worthwhile! >>>>>>>>>> >>>>>>>>> >>>>>>>>>Well deserved, Graham! >>>>>>>>>Thanks again for your work. >>>>>>>>> >>>>>>>>>[...] >>>>>>>>> >>>>>>>>>This game clearly showed that Fritz plays in a different league than Crafty! In >>>>>>>>>fact I think this was one of the best games of the WMCCC. >>>>>>>>> >>>>>>>>>Greetings >>>>>>>>>Alex >>>>>>>> >>>>>>>> >>>>>>>>My take on this game is a bit different. I do _not_ want my program to make >>>>>>>>such a sacrifice and then see the eval steadily go _down_ over the next few >>>>>>>>moves. It means one of two things for it to win such a game: >>>>>>>> >>>>>>>>1. The eval is bogus. It is saying "this is bad" when in reality "this is >>>>>>>>good". I don't want that sort of evaluation. >>>>>>> >>>>>>>But this is unavoidable. Otherwise computer programs would only need to do a 1 >>>>>>>ply search. >>>>>>> >>>>>>>> >>>>>>>>2. The program was lucky. A little luck doesn't hurt. But it doesn't win >>>>>>>>tournaments very often. >>>>>>> >>>>>>>Again, unavoidable. Have crafty play against itself and you will still have >>>>>>>decisive games. The games are won due to luck, since they have the same eval. >>>>>>>The question is, "did Fritz make a good gamble?" >>>>>>> >>>>>>>> >>>>>>>>Either the eval was wrong, or it was lucky. Neither one leave me feeling like >>>>>>>>"fritz is in a different league than Crafty..." >>>>>>> >>>>>>>Of course, but that is pretty much how _all_ games are decided isn't it? >>>>>> >>>>>>No >>>>>> >>>>>>There are games when one side get advantage and slowly increase the advantage >>>>>>without having a worse position. >>>>> >>>>>The only truly correct evals are a: win, draw or loss. The other stuff in >>>>>between are _practical_ assessments that do not correspond to the true >>>>>evaluation of the position, but they are precisely what all programs rely on in >>>>>all games. Yes? >>>>> >>>>>> >>>>>>> >>>>>>>> >>>>>>>>I suspect white has better moves that might have justified the pessimistic eval >>>>>>>>Fritz had... The right program might have made that sacrifice look as ugly as >>>>>>>>this game made it look brilliant... >>>>>>> >>>>>>>Better moves may exist, but you have to _find_ them. >>>>>> >>>>>>Crafty could find Nxe6. >>>>> >>>>>If Nxe6 is an improvement for crafty, it had to find it during the game and not >>>>>after. Why it didn't is irrelevant to the result. The result still stands. >>>> >>>>The result stands but the impression that fritz is a different league than >>>>crafty does not stand. >>>> >>>>Uri >>> >>>I have both programs. It stands, has been that way for a long time ! Fritz found >>>a move that Crafty could not find an answer for, all of the other stuff is >>>excuse making ! >>>Wayne >> >> >>I'm not trying to make _any_ excuses. Crafty lost. That happens. The issue >>(to me, now) is simply "did it _have_ to lose that game, was the sac sound, >>if not, why didn't it find the right response?" >> >>I always analyze losses to see what went wrong, otherwise there would be no >>way to make it play better. There are two ideas here: (1) if it should have >>found Nxe6 but didn't, then that changes things a lot. IE it shouldn't have >>lost but did due to operator error, my error, or a programming problem. (2) if >>it couldn't find Nxe6 on the hardware it had, period, then the discussion is >>now not about Crafty, but about Fritz, since it played a bad move but the >>opponent didn't punish it correctly. In that case, Fritz needs some tuning as >>it won't always get away with playing such a sac. There is no sense in a >>program impaling itself on its own sword... > >We cannot know if the evaluation of Fritz is wrong because we do not have the >source code of fritz. > >It is possible that fritz evaluates kings attack better and in this case there >was a king attack for the pawns. > >The fact that in this case the tactical move Nxe6 is probably winning does not >prove that the general idea behind fritz's evaluation is wrong. > >It is possible that Fritz also evaluate bad trades in similiar way to Crafty but >evaluate king safety in a different way. > >The discussion was about Crafty's moves. >I understand now that Crafty could not find Nxe6 without a good alpha even >without problems in time using. > >Uri I'm not even sure it could find this on a "good" alpha. I can't compare my quad to a single cpu machine with any degree of reliability for just one move. The single-cpu version of Cray Blitz finds this in about 1 minute, but in this position, there are several 'singular' moves that help. The 32 cpu machine would probably find it in a couple of seconds or so. But without singular extensions, it is a tough move to find as there are so many checks and things to blow the tree up. Wish Bruce would try Ferret on it to see how his SE implementation does on this position. It is also likely that null-move is not helping here either. This is the perfect kind of position for hidden zug positions.
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.