Author: Jonas Cohonas
Date: 13:16:23 04/25/01
Go up one level in this thread
Deep Fritz has finally gotten on the scoreboard! After five consecutive losses, Fritz sacrificed a rook to grab a perpetual check draw with black in game six. Junior's ability to sacrifice material to get at its opponent's king has dominated the match so far. There are still 18 games to be played, but based on the games (and Junior was attacking yet again in game six), programmers Amir Ban and Shay Bushinsky of Israel have good reason to be both proud and optimistic. Game seven is well underway and is looking drawish. Junior again sacrificed a pawn, but Deep Fritz defended well and managed to castle. Deep Junior got the pawn back in the endgame and it looks headed for a draw. 1 2 3 4 5 6 7 8 9 10 11 12 Score Deep Fritz 7 0 0 0 0 0 ½ 0.5 Deep Junior 1 1 1 1 1 ½ 5.5 Some technical notes sent from match director Professor Enrique Irazoqui: 1. Fritz 7 is also "Deep" (utilizing dual processors). 2. Fritz's NPS (Nodes [roughly positions - ed.] per second) is a little less than one million, which is slower than the Fritz 6 version everybody is used to. 3. After game five Prof. Irazoqui swapped machines, transferring both programs and the books with the learning values. [To ensure no advantage, much the way teams switch sides of the court in most team sports. -ed.] 4. Enrique also measured the performance of both machines and found them equal (0.2% difference). 5. He also managed to repeat both programs' moves on a different machine. 6. He expects to file more results in 14 hours. The winner of this match, scheduled for 24 games, will play world champion Vladimir Kramnik in October. All results and games will be posted here with updates of games in progress when possible. You can download the game with full log files (with the machine's evaluations after each move) in ChessBase format. KC thanks match director Prof. Enrique Irazoqui for his help. You can follow GM Boris Alterman's commentary move by move in the viewer or download it in PGN, links on right. The numbers near each move represent how many thousands of positions Deep Junior looks at per second, its evaluation of the position (positive for white, negative for black), depth of thinking in half-moves, and time spent on the move. In game one Fritz was up a pawn but failed to realize that black had enough compensation. When it tried to play for the win, Deep Junior's bishops made it pay. 1-0. Game two illustrated the problems Fritz has had. An exchange sacrifice gave Junior a vicious attack against Fritz's king. From looking at the logs, it's not that Junior saw the attack all the way through, but that it simply has a higher (and more accurate) evaluation of open chances against its opponent's king. 2-0. Game three was a combination of the first two games in many ways. Junior went down a pawn after enthusiastically pushing it forward, but gained attacking tempi when Fritz stopped to collect the material. Then DJ sacrificed a bishop to begin a merciless series of checks against Fritz's king. A remarkable sequence of close to 20 consecutive checks (!) produced the win of white's king and soon after the game. 3-0. Game four looks to me like a book problem, at least for a computer. They played a topical line of the Petroff Defense in which black sacrifices the exchange for compensation in activity and a passed central pawn. (It turns out that DJ was in its book for a few moves longer, so Fritz sacrificed the exchange on its own. But there really wasn't much choice at that point.) Unfortunately for Fritz, computers are merciless at exploiting material advantage and playing into this position with black in computer play is close to suicide. Notable is that the less materialistic Junior rated things as about even after the opening, but Fritz hated its own position! The queens came off, the grind was on, and DJ soon had an extra rook and another full point. One hilarious only-in-a-computer-match scene arrived when white had two rooks and a pawn versus black's lone rook. A rook advantage is only +5 in computer-thought. But since both programs are using endgame tablebases, which enable them to play any position with fewer than six pieces (including kings) perfectly, it can see that the rook and pawn versus rook endgame is forced mate! So for several comical moves, Junior tried to throw away its extra rook and Fritz refused to take it! Finally Fritz took the rook and was duly mated in 34... 4-0 Game five was another wonderfully speculative king attack by Deep Junior. It is remarkable that this sort of play can succeed against a calculating beast like Fritz, but DJ's "intuition" seems to always know when such an attack will pay off in the long run. A queen on the open board in the hands of a computer is a terrible thing, and although Fritz (and, to a lesser extent, Junior) saw the position as a draw for a long time, Junior slowly made progress. Finally the white pieces could not maintain coordination and the white king was executed. 5-0 Regards Jonas
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.