Prophet @ ACCA World Computer Rapid Chess Championships 2008

Last modified: 06/24/08

Prophet participated in The 2008 Second Annual World Computer Rapid Chess Championships on June 21st and 22nd, 2008. The tournament was an online tournament, hosted by the Internet Chess Club (ICC). Prophet was seeded 17th out of 36 entrants (I think a couple late entries came in after the seedings were posted though). I expected to get around 6 or 7 points, maybe a little more with some luck. Prophet finished with 6 points. Not too bad, but I hoped for a little more than a 29/36 finish... just a tad disappointing.

In terms of the evaluation function, Prophet really hasn't improved any since the last tournament. And the search hasn't improved too much either. My focus has been on implementing Dynamic Tree Splitting (which is incomplete as I write this), so I wasn't able to use that. However, Prophet is running on pretty good hardware this time around - a Core Quad Q6600 2.4 ghz overclocked to 3.1 ghz, with 2gb ram, running 64 bit Gentoo Linux. For the first time in a tournament, Prophet is running with 4 search threads (using YBW since DTS is not ready).

I don't have the time right now to analyze all these games - there were 14 rounds. For now I will just post the pgn and log files and a few comments about the experience. I do want to come back and analyze a few though - the games against Romichess, Symbolic, Tinker and YACE.

Round 1 - Prophet vs Neurosis [PGN] [LOG]0-1

Round 2 - Romichess vs Prophet [PGN] [LOG]1-0

Romi was being operated by Tony Thomas. Something was definitely off with Romi. It was showing wack scores and the time management was horrible - it nearly ran out of time. Prophet was up and looked like it would win. Near the end of the game, with less than a minute on Romi's clock, Tony disconnected and reconnected with a "tried and true" version of Romi (as he put it). Somehow Prophet gave the game away. That was pretty embarassing!

Later the TD (Charles Roberson) reversed the game because of the rules infraction by Tony - you're not supposed to disconnect and reconnect with another engine - even a different build of the same engine. I personally didn't mind, but I understand the decision. However, I'm not claiming the win, despite what the official tournament results show. [Update: re-reversed? the cross table on the ACCA site shows Romi won.]

Round 3 - Prophet vs Tinker [PGN] [LOG]0-1

Tinker is a pretty solid engine. I've met Brian Richardson, the author, on a couple of occassions - a real nice guy. He told me he's been doing a lot of work on Tinker lately - primarily squashing bugs and cleaning things up. I don't mind losing to Tinker, but it looked like Prophet was winning near the end. Another blown ending?

Round 4 - Roce vs Prophet [PGN] [LOG]0-1

Round 5 - Prophet vs HFC [PGN] [LOG]1-0

Round 6 - ZCT vs Prophet [PGN] [LOG]1-0

ZCT is new to the tournament scene I think. I had a nice chat with the author, Zach Wegner, about parallel searching. His engine is one of the very few that already uses DTS.

Round 7 - Prophet vs Tornado [PGN] [LOG]1-0

Round 8 - Symbolic vs Prophet [PGN] [LOG]1-0

Ahh... good old Symbolic. I didn't have the same luck with Symbolic that I had at the 2006 ACCA Closed. I haven't analyzed this game yet, but I seem to remember a rather risky looking sacrifice by Prophet to gain a king attack (that obviously didn't work out).

Round 9 - Prophet vs Deltomatex [PGN] [LOG]1/2-1/2

A rather solid game that appropriately ended in a draw.

Round 10 - Buzz vs Prophet [PGN] [LOG]1-0

I've probably run more test games against Buzz than any other engine - thousands even. Prophet gets around 45% on my test setup. This wasn't one of those.

Round 11 - Prophet vs Horizon [PGN] [LOG]1-0

Got this one, but it was close for a while.

Round 12 - Noonianchess vs Prophet [PGN] [LOG]0-1

One of Charles Roberson's engines. Telepath is the other (much stronger) engine. Prophet simply outsearched Noonianchess.

Round 13 - Prophet vs Petir [PGN] [LOG]1/2-1/2

A rather lucky draw out of book.

Round 14 - Prophet vs Yace [PGN] [LOG]0-1

I didn't expect to win this one so I didn't pay much attention until the end when I saw a kibitz from the operator (I'm not sure who the operator was at the time). It looked like Prophet had a chance going into the endgame, but it didn't work out. This game would be worth analyzing as well... another case of underestimating the opponent's advanced pawns.
After every tournament I like to kind of assess things; to think about what needs to be done to improve before the next tournament. For now my main focus will continue to be on implementing DTS. I really don't expect DTS to be significantly better than YBW (which I already have), especially on a machine with just a few cores. My intuition is it will be more efficient on machines with many cores, but how much more efficient remains to be seen. So really, DTS is really more of a long term investment that may not even pay off! But it's an interesting project.

What would help Prophet over the coming months is for me to get more serious about improving its evaluation function. I analyzed several games with Crafty in the Dasher interface while watching them. Prophet's eval was just too far out in left field in too many positions. The evaluation function is by far the biggest weakness in the program. So I am going to start (again) with a disciplined testing program using a handful of opponents.

Finally (well not finally, but all I'm going to write about), Prophet is giving away too much time in the opening. It's getting outbooked. The opening book was constructed with a PGN file supplied by Dann Corbit (I think). The PGN file is made up of many high quality computer games. I'm not seeing any outright blunders out of book, but the book is shallow. Prophet routinely gives up as much as five minutes in a 25 4 game!
Back to the main page