Gomocup 2022, the 23rd tournament (April 15th-17th, 2022)

The Gomocup 2022 took place on April the 15th-17th, sponsored by Tech Center of Lightspeed & Quantum Studios Group, Tencent.

There were three cloud servers whose configurations were the same:

  • Tencent Cloud Server SA2 (Windows Server 2019 Datacenter, x64 32 vCPU (AMD EPYC ROME, 2.6/3.3GHz), 64GB RAM)

The openings for Gomocup 2022 were chosen by the following people (sorted alphabetically according to last names):

  • Alexander Bogatirev - Gomoku player, member of Gomoku Committee RIF, member of organizers committee of Russian Gomoku Championship, winner of Russian Gomoku Cup 2016.
  • Tomoharu Nakayama - Japanese renju player, 9dan. He got the 7th place in the 15th World Renju Championship, and won the Japanese Meijinsen in 2017.
  • Aivo Oll - Renju player, 7 dan, general secretary of Renju International Federation, former Estonian champion, European champion, and world champion.
  • Zijun Shu - Gomoku player and AI researcher, contributor of several Gomoku AIs.
  • Tao Tao - Renju theory researcher, who published some researches and new designs of renju opennings, participated in promotion of Renju and translated several Japanese renju books.
  • Qichao Wang - Renju expert, 4 dan, who has done a lot of researches in renju openings, and won the World Correspondence Renju Championship 2019.
  • Rong Xiao - Gomoku expert who proposed the gomoku opening rule "swap after first move", which is one of the most popular gomoku opening rules in China.

Thank you all!

Same as last year, we prepared 12 openings for every rule.

  • Openings for the freestyle league:

  • Openings for the standard league:

  • Openings for the renju league:

What is new?

  • Updates
    • AlphaGomoku (Kozarzewski) 2022 - Now using Monte Carlo Graph Search. Fixes and optimizations in the VCF solver. Improved time management. Added several utilities like benchmarking or automatic configuration.
    • Embryo 2022 - The evaluation also takes into account 3 + 2 and 2 + 2 patterns. C++ code ready to run on Android, release with GUI coming soon.
    • PentaZenNN 2022 - Switch to MCTS with NN evaluation modified from KataGo. The alpha-beta search serves as a parallel surewin finder now. PentaZenNN supports standard and renju rules.
    • Rapfi 2022 - Rapfi is now equipped with an NNUE (Efficiently updatable neural network) trained on games played by Katagomo. Although NNUE runs with a lower speed, its accuracy on evaluation brings notable strength improvement. Classical evaluation is improved via automatic tuning, and combines with NNUE in a hybrid way. Search and speed is also improved.
  • New AIs
    • Angel - a Gomoku program developed by Yuan Xie. Negamax Alpha-Beta search, possible position sort, quick believable VCF search, Iterative Deepening. Angel supports freestyle rule.
    • Arbillo - a Gomoku program developed by Lorenzo Piazzo. First attempt. Standard (alpha-beta) player. Arbillo supports freestyle rule.
    • DDQK-Conquer - a Gomoku program developed by Sui Shunyi. Mainly using the use-value network algorithm, alpha beta screening is also involved. DDQK-Conquer supports freestyle rule.
    • Zhoumian - a Gomoku program developed by Shangren Shang. A low-IQ AI based on C language. Zhoumian supports freestyle rule.

There were 2 freestyle groups (i.e., Freestyle 1 and 2), 1 fastgame group (Fastgame), 2 standard groups (i.e., Standard 1 and 2), and 1 renju group (Renju) in Gomocup 2022. Elo rating system was used to evaluate AIs' strength. AIs were divided into different freestyle and standard groups according to the placement in the last tournament. For Freestyle 2 and Standard 2, the top 4 AIs were moved up to the next group (i.e., Freestyle 1 and Standard 1, respectively). If the top k (k>4) places were all taken by new (or updated) AIs in a group, then all these k AIs would advance to the next group.

Memory limit/time per move/per match were determined to be same as Gomocup 2021:

Freestyle 1 league 300 1000 350 20 five or more stones
Freestyle 2 league 30 180 350 20 five or more stones
Fastgame 5 120 350 20 five or more stones
Standard 1 league 300 1000 350 15 exactly five stones
Standard 2 league 30 180 350 15 exactly five stones
Renju 300 1000 350 15 renju rule

In addition, following the success of the previous two experimental tournaments where AIs were running on participants' computers, we continued organizing a tournament of such kind and officially named it the "unlimited tournament". The unlimited tournament had no limitation on AI size and allowed participants to use any hardware (e.g., GPUs) that they had access to. Time per match was 120 min. There was no additional time limitation per move. Following Gomoku World Championships, the rule for this tournament was Swap2. The tournament was a double round-robin tournament and tie-break criteria were: game points; match points; Berger coefficient; personal score, following World Correspondence Gomoku Championships.

There were technical difficulties like every year.

  • The tournament got stuck multiple times. We noticed that at the point when the tournament got stuck, AlphaGomoku was always one of the active players, and the crashes were all similar, with bad_alloc exception shown in the stdout/stderr of AlphaGomoku. The crash of AlphaGomoku further stuck GomocupJudge. To unblock the tournament, we added in GomocupJudge a special restriction to the number of running AlphaGomoku instances and had the tournament resumed.
  • Multiple AIs in the unlimited tournament had connection issues due to bugs in their protocol implementations. The tournament was paused until the bugs were fixed.

WINE, SLOWRENJU, HEWER, and TITO were the top 4 in the freestyle 2 league and advanced to freestyle 1. ARBILLO was the best-performing brand new AIs.

This year the top 3 AIs in the freestyle league all leveraged neural networks. The updated RAPFI became the new winner of the freestyle league this year. KATAGOMO took the second position, and BARBAKAN was the third.

The winner of the fastgame league was RAPFI. The second place was taken by BARBAKAN, and the third by EMBRYO.

TITO, HEWER, SLOWRENJU, and WHOSE were the top 4 in the standard 2 league and advanced to standard 1.

Similar to the freestyle league, the top 3 this year in the standard league all leveraged neural networks. RAPFI became the winner of the standard league. The second place was taken by KATAGOMO, and the third by ALPHAGOMOKU (Kozarzewski).

The winner of the renju league was RAPFI. KATAGOMO took the second place. The third was EMBRYO.

Four AIs participated in the unlimited league. The winner of this league was RAPFI. KATAGOMO took the second place. The third was ALPHAGOMOKU (Kozarzewski).

Congratulations again to RAPFI for winning all leagues!

You can download complete results and openings here.