Author: Vincent Diepeveen
Date: 09:43:56 10/30/01
Go up one level in this thread
On October 29, 2001 at 17:26:43, Gian-Carlo Pascutto wrote: >On October 29, 2001 at 16:42:32, Vincent Diepeveen wrote: > >>Also don't forget that this championship is 90 0. All games on the >>internet where crafty is tuned for are against usually humans and >>way faster levels. > >I have been seeing similar problems. At 90 0 it's seldom >to see tactical shots. You either get positionally outsearched >or outevalled. > >>In diep it seems also something is real bad tuned for slow level. I >>didn't even manage to win from a program called Spiderchess (though in >>analysis it appeared it with some luck played the openings setup pretty >>solid). > >I am really impressed with Spiderchess. Don't forget they also drew >ChessTiger. They literally walked over Sjeng. He was pretty lucky so far getting closed positions. It'll get annihilated in open positions though. In closed positions the 50% chance rule is there. Definitely a good result. Big compliments to spiderchess. He couldn't complain about stupidity from DIEP's side btw. In fact the mistake was to get a KID a look like position from DIEP with features which were with some better analysis GOOD for black. DIEP just played some strategical horrible moves just to get a strong bishop versus a weak bishop. The real problem for spiderchess is getting above that 50%. We've seen many engines like it in the past, most of them were directly scoring pretty well, but none of them came much further than that, because as soon as you try to fix the problem: adding knowledge, then obviously the style of the program changes to a style which all the others already have. If you play like me, but with less knowledge, then i completely rape you. That will happen to spiderchess against similar type of programs real soon. In a closed position you then you get a draw quickly. I remember how much problems i had in the past with Patzer. Recently some knowledge was added to it though, so what diep finds a cool plan to play for, is more likely now something which patzer likes to play for than it was in the past (that's my impression, correct me if i'm wrong) and if i test now against the winboardII (or whatever) patzer then i score an amazing 90% against it now, whereas the older versions in tournaments always did a good job against me in closed positions, just like Spiderchess did last weekend. >I got out with a cramped game, and they nicely alternated >between attacking both the king and queenside. Sjeng helped >them with h5??? and blew a drawing chance in the endgame too. The advantage of a simple engine is that you can debug it very easily. The problems come when you have a well debugged engine and want to improve it. >The program is very well tested. Also amazing is that they don't >nullmove! Definitely well tested, but on average everyone is way better tested. See how EEC and Gadget walk into the tournament. I remember how DIEP 1.4 was when i joined dutch open. It didn't have anything. No hashtable, no nullmove, no pondering, and it would never stop playing until i resigned or mated my opponent. It even accepted illegal moves, causing random moves to happen on the board after it had captured the king internally. Gadget would crush it!! >>Everyone knew in advance tiger would win. > >Met zo'n instelling komt daar dan inderdaad ook nooit >verandering in. Kneus ;) I can't do anything against business decisions from others. If 2 programs from the same company join with the same book, then there is someone always who decides without it being clear that one is going to win and the other is going to lose. In this case let's say the openingsscore from both engines was around zero :) >>Faster hardware would not have helped there. The seemingly silly queen >>moves from Crafty against DIEP are not something which a deeper serach >>will solve. All crafty tries to prevent is partly delay DIEP from >>castling and in fact DIEP didn't mind that at all initially. For that >>delay DIEP got itself a superb pawn structure. >I was praying you would rape crafty in a kingside attack. IMHO >Crafty's kingsafety code is a joke, and I'm hoping someday Bob >realizes that and rewrites it. I could use some new ideas for >mine :) It was pawn structure which killed crafty this time. Only when it was too late then crafty started regretting its position. Happens. I hope Crafty scores well now against the 4.5 point programs. Losing against Isichess was definitely not needed. Rg3?? was a major tactical mistake against Isichess from crafty. Rh3 there is the move to play. Now diep doesn't find Rh3 either itself, but at least it has the right idea to attack: Qg4 followed by Rh3 is what diep is going to play. Now that loses a crucial tempo, but it still puts major pressure onto blacks king position, knowing that most programs lose tempi anytime of the game anyway. Crafty completely misses this tactical pressure because it doesn't evaluate anything that looks like it. Nevertheless it was a poor book job by white. Playing a line that's known to be lost for white, playing it with a tactical weak engine, and in the last place the first random book move from Isichess already confusing crafty completely. Sadly also the automatic generated book from crafty didn't help it here. Nearly all books that play with white Italian will have 18.Rh3 there. Even more amazing than that Crafty lost it, is that deep junior 7 says white is up 0.76 pawns here (it's planning qg4 followed by Rh3 here too). After half an hour that goes even up to +0.96 for white for Qg4 followed by Rh3. Really amazing. DIEP's more realistic here saying black is up 0.106 and later this goes down even. >-- >GCP
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.