Shay Bushinsky: how Deep Junior conquered Turin

How Deep Junior conquered Turin

By Shay Bushinsky

The beginning of this millennium coincided with a rugged race between two of the best chess engines of the World: Shredder and Deep Junior. In the world computer chess championships held in London in 2000, Shredder won. Then came two successive wins by Deep Junior: 2001 and 2002 (both in Maastricht) while Shredder regained the crown in 2003 (in Graz). Deep Junior fought back and won the title on its home turf in 2004 (in Ramat-Gan).

The world Computer chess title was thus portrayed as a private affair between the accurate and solid Shredder vs. the more daring and creative Deep Junior. However the 2005 championships held in Reykjavik changed it all. Out of the blue came two unknown programs – Zappa and Fruit – which took the top two prizes.

Computer Chess benefited from the appearance of both programs, but especially from Fruit, since its author decided to publish it as open source. As a result, a flood of new very strong programs appeared on the computer chess scene. One of them, Rybka, caught everyone’s eye by demonstrating extremely strong chess.

The 2006 World Computer Chess Championships was scheduled to be held in Turin, Italy. For the first time, the tournament was staged in conjunction with the human chess Olympiad. The chief organizer, Prof. Paolo Ciancarini of the University of Bologna, succeeded in blending the world’s best chess players together with the best machines, thereby creating unique opportunities to view both in action.

The computer chess world championship in Turin

The rather short time between the 2005 championships (held in August, 2005) and the Turin event (in May, 2006) made our preparations much more difficult. After Reykjavik, we understood that fundamental changes must be made in Deep Junior’s code including its search algorithm and evaluation function. The emerging new engines seemed to defeat the new effort while discovering on a monthly basis a new and stronger opponent. The testing procedure of the new Deep Junior also was a problem. While Shredder’s latest versions seemed to be our frame of reference, many more programs had to be pitted against Deep Junior. The commercialization of Rybka frequently drove new and strong versions. It seemed to have a huge following with testers and opening book authors.

Nevertheless, even though it seemed at the time that no new search techniques could be introduced to Deep Junior, some ultimately worked. The new Deep Junior was almost 100 points better than our winning version of 2004, which is certainly a significant leap. Similar reports came from Shredder, who introduced his new version just a week before the championships. Rybka was already reported as surpassing Kasparov’s record of 2850 Elo points with new and innovative search techniques of its own.

The winning team in action in Turin: Amir Ban (left) and Shay Bushinsky (right)

Another factor was the machine that Deep Junior would be running on. This time we were offered by Intel to try out their new Woodcrest, a dual duo core machine (four cores in total). This machine had not been released yet and we had little time to adapt the code to it. Our initial testing showed satisfying results, and after employing optimization of the Deep Junior code we peaked about 9.3 million nodes per second, which was about 33 per cent faster than the platform we used in Reykjavik.

But then the news came that the 2005 Champion, Zappa, would be using 512 processors on the NCSA super computer. The net gain for Zappa was a peak of 100 million nodes per second, and its author reported a 3-4 ply deeper search than the 2005 Zappa, which sounded threatening.

Under such circumstances, we arrived in Turin with few expectations. In previous championships, Deep Junior was notorious for dropping points to weak opponents in the opening round. But this time it started strong and scored three wins out of three games after two days of competition, which was an excellent sign that the version is doing fine.

A key game between Rajlich (Rybka) and Shredder

This time it was Zappa that dropped an unexpected half point to Jonny on the first round. But Shredder showed its real progress by defeating Rybka in the third round quite comfortably. That meant that just before the fourth round we and Shredder were the only programs to score 100 per cent, and we had to meet each other on the next round.

The key match with Shredder was of course exciting. Both programs already had played each other several times and this always meant fireworks. In a very sharp Sicilian position Shredder made a slight error, giving Deep Junior an outside passing pawn with what looked to be serious winning chances. Shredder, notorious for its persistent defense, managed to tame Deep Junior and the game ended in a draw leaving both programs in the lead. Yet Shredder was happier because it had already played Rybka, and we didn’t so it had been positioned better to win the title.

Amir Ban and Shay Bushinsky during the key game Deep Junior vs. Shredder

In round 5 nothing much changed. Shredder gained more of an edge by drawing with Zappa while we drew with Diep in a game we normally should have won.

In round 6 we played a key game with Rybka which ended in a draw. But the good news was that Jonny drew Shredder so we still shared first place. Rybka and Zappa were both breathing down our necks with just a half a point behind us.

In round 7 we faced Zappa in another critical game. Zappa had to win. Otherwise, it would remain behind with both Shredder and Deep Junior surpassing the strong opposition. After a bad opening with the black pieces Deep Junior found itself for the first time in the tournament in a really inferior position lacking a pawn with no eminent compensation. Just when it looked as if Zappa was about to take the lead, Deep Junior made a bold move that provoked its massive opponent to capture material. Soon enough Zappa found itself in an awkward position needing to force a draw otherwise it would be mated by Deep Junior.

The bad news was that Shredder became the sole tournament leader after defeating Diep. With four rounds to go and Shredder looked unstoppable without any threatening opponents left. We had to win against Jonny, who previously managed to draw out of all the top programs. Deep Junior produced an excellent positional game and outplayed its strong opponent. This time it was Shredder’s nemesis, Spike, which managed to grab a half a point from the leader.

Thus after 8 rounds we equalized again with Shredder and the smell of playoffs was in the air. The regulation of the tournament states that if two programs end up with equal points they must play two playoff games to decide the winner. We were in a playoff situation with Shredder in Maastricht 2002 and prevailed. Yet everybody wished to avoid playoffs which would steal time from the closing ceremony and prize giving dinner.

Round 9 didn’t change much with both Shredder and Deep Junior winning. For us this win meant a sweet vendetta over Crafty, which managed to defeat us in Reykjavik. But the real surprise happened in round 10 when Shredder had “an easy” game vs. ParSOS, its compatriot program. After a bad opening, Shredder conceded another draw. For the first time we were in the lead alone and well positioned to win the championships provided we win the last round.

Round 11 summoned for Deep Junior the last strong opponent – IsiChess. It already extracted a draw from Deep Junior back in 2004. Therefore, the hopes of all other contenders were that with the help of the white pieces history will repeat itself.

In a nerve-wracking game IsiChess looked good, managing to exchange much of Deep Junior’s gun fire. A position was reached where Deep Junior edged a middle passing pawn to the end of the board. But the pawn was blocked by IsiChess rooks, and it remained to be seen how Deep Junior could further its position. Deep Junior found the correct plan, trying to penetrate with its king to assist the passing pawn, and IsiChess was forced to error and create another passing pawn for Deep Junior. Passing pawns are notorious for being difficult to evaluate as their progress is very slow and only after very deep search they could be proven efficient. Deep Junior won the game and the thus became the 14th world computer chess champion!

A couple of famous grandmasters stopped by to play with Deep Junior in Turin. In this picture chess GM Ljubomir Ljubojevic (famous from the Bobby Fischer era) tests Deep Junior in a blitz match. Sitting next to me (right) is Prof. Jaap Van der Herik, the computer world championships tournament director. Next to him Dutch chess aficionado and philanthropist Joop van Oosterom and Dutch grandmaster Jeroen Piket.

Naturally, both Amir and I were happy with Deep Junior’s performance. Not only did it live up to its reputation by sacrificing material in every game for the initiative and arguably playing the most entertaining chess, but also its evaluations during the tournament where very stable and proved correct. Some games had elements of creativity which always give pleasure to the programmers.

• Click to replay all Deep Junior games in Turin

Final standings

• Official web site of the Computer Chess World Championship
• Individual round-by-round results and cross tables
• Games and bulletins

Junior 10 is now available in the ChessBase Shop. It provides you with twelve months of free access to the Playchess.com server. You also get the latest openings book which is continually enhanced and improved by GM Boris Alterman, a member of the Junior team. Deep Junior 10 is a special version that runs on multi-processor systems, making it much faster than the single-processor Junior 10.

Order it now: