Freestyle Chess –Teaching an Engine how to Fly

The Flying Saucers team consisted of one human player + computer(s). I am 30 years old and a mathematician by education. This was my third Freestyle event. I have played as a sole centaur each time. Currently, my main activity in chess is freestyle chess and opening studies with this in mind. I am also a member of a local chess club (2163 FIDE Elo), and enjoy the socializing that takes place around low-level club league matches.

My hardware in the preliminary was an AMD dual core Opteron, which has to be considered sub-average hardware these days. As it is, I know for certain that at least one other finalist qualified with even worse hardware than me, so to get that question out of the way, no, Freestyle chess is still not only a question of spending bucks on hardware. My three wins in the preliminary consisted of one "endgame-swindle", an attractive attacking game vs. GM Yuri Solodovnichenko (who, despite this loss, managed to qualify for the final), and finally a successful defense against a wild Bxf7+ sacrifice in the 2.c3 Sicilian.

Before the final, Rybka programmer Vasik Rajlich had once again kindly offered me to let me use his best system, this time a quad-core Intel. My technical setup then was: Fritz 9 interface, with two instances of Rybka running, the quad on internet-pipe running in 1-variation mode, and my own dual core running in 2/3/4-variation mode. This way, I would be alarmed by deep resources and assessments quickly, while at the same time getting immediate information about the forcedness of the investigated positions, in other words, how many alternatives would also be worth a check.

I decided to use only one engine this time. The advantage of using several engines is that different engines may evaluate positions differently, and this (+search differences) results in one engine finding some potentially strong move faster than another engine, and vice-versa. However, and I think this is an important point, a key part in strong centaur analysis is to know both the strong and the weak points of your favourite engine. The engine output needs to be interpreted, and here experience with your engine is very important: What kind of lines does the engine tend to not pay enough attention to, and which types of positions does it tend to misevaluate. If you have a good feeling for this and have developed methods to adjust for it and integrate your own input, you should have a considerable advantage. Once this is achieved, I am not so sure the output of a second or third engine will improve the decision process much. At any rate, I didn't have the hardware to support a second engine, nor the (necessary) experience with working with two different engines at once.

Unlike last time, I used the endgame TBs installed on my system, which indeed turned out to be a probably better decision. The games themselves, I played through the free Playchess client.

After the preliminary play-off matches, I once again spent considerable time preparing for the individual opponents in the final based on the expected pairings. However, four days before the final, a random pairing was announced, and I had to scrap much of my preparation, and also had to settle on only getting four whites. Nevertheless, one thing was still clear: The final could very well become some kind of a Najdorf festival! A majority of finalists were 1.e4 players, and I think also a majority were black Najdorf players, at least potentially. This suited me quite well, as I had been intensely studying various topical lines for some months prior to the final.

As it turned out, I ended up in three Najdorfs as black, and one Najdorf-turning-into-an-odd-Rauzer as white, and those four games gave me the number of wins necessary to clinch first prize. I also had some pressure against runner-up team Cato the Younger after they had gambled with 1.h3 (intending a French in the forehand), while I was under quite serious pressure as black against Kaputtze in round seven and as white in round nine vs. Ciron, a very nervous affair on my side. My remaining two white games were fairly uneventful draws. Perhaps next time, I will insist on getting nine blacks in the final!

Overall, I am of course very satisfied with my performance. I managed to hit the ball when I got the chance, and I held the draw in the couple of games where I got the worse of it.

I have annotated five games, two from the preliminary, and three from the final. I have mainly focused on situations where the "human touch" was crucial, trying to relay what went on behind the scene. Various decision methods and issues of pre-game preparation are touched upon. Finally, I hope that the reader will also simply find the games themselves interesting and fascinating!

Anson Williams, 27, was the team leader. A veteran centaur player, he is a Telecommuncations Engineer / Software developer, and skilled with computer hardware. He is an unrated OTB chess player. As “Intagrand” he has placed 3rd in the 4th Freestyle tournament, his team consisting of the same cast of characters. Among his interests are bowling and J.S. Bach, which he plays about as well as he Freestyles. A graduate of Imperial College with a Masters degree in Electrical and Electronic Engineering, Anson now lives in London.

Yingheng Chen, 24, a graduate of The London School of Economics (LSE). She assisted Anson with analysis during all nine games of the finals. Her interests also include bowling, badminton, cycling and reading. Through much exposure to computer chess (solely because of Anson) she has picked up some useful analysis skills and was a great asset during the tournament.

Nelson Hernandez, 50, lives in the United States and did not participate in the games as an active player. However he has collaborated for years in developing the team’s book, and with Anson is a tournament co-strategist. A former paratrooper and stockbroker, he is now a financial analyst. He is also an unrated player, and says he hasn’t played a competitive game of chess OTB in over 20 years. The team name of “Cato” comes from his admiration of the principled Roman Senator who vainly opposed Caesar and finally fell on his sword.

Anson and Yingheng had several computers running during the tournament of varying power. His top-end machine was a quad core Intel QX6700. Beyond that, he considers the specifications of his machines and the functions they performed “classified”.

At different points in the tournament different chess engines were consulted, though not all at the same time. The specific engines that were used are likewise “classified” but any well-informed person could make an educated guess and not be far off the mark.

The team has absolutely no comment other than to say that their book was not a decisive element in any game of the qualifiers or finals, though it surely saved the team from countless traps they might have fallen into against such formidable opponents. It generally served its purpose: to reach the middle-game with an even or slightly better position on the board, giving Anson a chance to work his centaur skills.”

We played qualification round under the Welsh flag. First, because the Welsh flag is lovely, second, to cover our identity and rule out any possibility of preparation based of former games in the final. Unfortunatelly, our identity was disclosed by Arno Nickel, who had information from organizers, so we played in the final under Czech flag. Our team consists of three people: Jiri Dufek (in the 4th Freestyle a captain of winning team Xakru), 2276 FIDE and strong ICCF IM. He is a computer expert and clearly one of the best freestylers I have ever played with. Then Miroslav Kvicala, 2042 FIDE, a keen fan of Scandinavian Defence, who has recently excelled also without engines, scoring 10 out of 11 in OTB chess league, with performance over 2300. The last person in the team was Patrik Schoupal (myself), a founder of a well-known Freestyle team Equidistance (this time without our favourite Morra Gambit) and a keen theoretician of chess openings.

Etaoin Shrdlu team member Patrik Schoupal

Engines used: Our main analytical engine was Rybka, but we used also Loop and for some games (positions) Junior and Shredder too. Also Hiarcs was running on infinite on one computer. All in all, about five engines, Rybka as a main analytical tool + other engines for ideas or even as a main engine for some specific positions.

Computers: I analysed on my E4300 Core Duo, overclocked to 2.67 Ghz. Mirek used dual proc. notebook and one more Core Duo computer connected remotely. Jiri played on about 5 remotely connected computers, ranging from Athlon dual to E6700 Core Duo. We hadn't a Quad processor (unfortunately!).

Miroslav Kvicala, 2042 FIDE

Repertoire: We were rather creative, unfortunately, not too much successfully creative, trying to outsmart our opponents in some games, but in the end, being outsmarted by them. E.g. in a game against Ciron 1.Nf3 Nf6 2.c4 e6 3.g3 d5 4.b3!? trying to avoid deeply analysed lines, just to realise a move later, that after 4…d4! we are getting slightly worse position. Or in the second round, we played against an engine-only player Hercules01 and I got an idea to copy the quick win from the first round game Rodo vs Flying Saucers, when Flying deviated in Najdorf with 24…Kh8 from well-known position and easily rolled over poor Rodo's Rybka engine. Unfortunately, our second round opponent had this line in the book, with the very strong 25.Rh3!, and we were just lucky to escape with a draw in the end.

In round 5, the exactly same line was played in Rodo vs The Wizard of Os. Rodo was not playing in engine mode, so he could easily follow our game, and when black wanted to "improve" with 26…Qxe4? he was of course crushed.

I am an average club player with a local (German) chess rating of about 1720 DWZ. I play at the best chess club in the world – Chaos Mannheim. Due to my time consuming job I have only little time for playing chess, except for late night online bullet games on Playchess.com and participation in online freestyle tournaments. I play about 6-8 over the board tournament games per year. (Jochen told me later on, he played together with two chessfriends, who are considerably stronger than him, but no titleholders. A.N.)

About book preparation he said in his answers to a questionnaire: We assembled some opening books which relied on a wealth of correspondence chess games. The preparation time was nearly nothing, since we constantly keep those books up to date.

Interesting is his theory about the high number of Najdorf games in the 5th Freestyle Tournament: The popularity of this opening, at least in freestyle and engine tournaments, might stem from the fact that people iteratively improve their books. Thus, the books tend to contain more and more variations of the Najdorf than on other openings, and thus the likelyhood to play the Najdorf in those games is higher. Me personally, I would like to see people exploring other openings with equal intensity.

And what about the phenomenon of two players with all their games drawn: Being one of the two ‘kings of draw’, I have thought a lot about this question. We played for a win without taking uncalculated risks. Both my team as well as Arno Nickel might have used books, derived from correspondence chess games. My theory is that it takes a lot of time to prepare a book, which is not only safe (granting a good play) but which also contains new, unpreceded killer variations.

I did not find the time yet to analyse all games of the tournament with a decisive result, but I theorize, that only rarely, a game was won out of an equal middle game – at least in the finals.

We are a team of a few strong Dutch players. We didn't have any experience in Freestyle chess, so it was completely new to us. The hardware we used: two dual core machines and one quad core, made available
through computer chess friends. We used mainly Rybka and Rybka Winfinder as analysis engines, sometimes switching to Shredder or Hiarcs. In the preliminaries we scored the necessary +3 to go for the tie-breaks. In the
tie-break we overcame PawnStriker with a 1,5-0,5 victory. In the final we managed a 50% score, which we see as a good achievement, considering the fact that we have no experience in this field. Thus we can look back to a good tournament, giving rise to do better next time. (They added three annotated games, which show their decision-making and reasoning.)

Night life in Abu Dhabi with the Hydra Chess Computer