This project is still under development
Next major steps:
- Identify the best setup to re-run all 960 x 20.000 games in SF17.
- Plot, for the standard opening and maybe one other opening, the white/black wins and draws for up to 50-100k games.
Identify optimal openings.
-
We analyze about 20 million Chess960 games generated with Stockfish 16.
-
We give a formal definition of an opening and apply it to the dataset.
-
We rediscover well-known chess openings and identify novel openings for all Chess960 variants.
Consult the project description for details.
- Amateur and professional Chess960 players.
- Anyone curious about the scientific analysis of Chess960 openings.
Here is the example for the standard opening and the opening with Queen and King interchanged, indexed by their Starting Position Index. We rediscover well-known openings and their variations, such as Queen's Gambit, Sicilian Defense, and the Ruy Lopez. Click on the board link for details.
| SPI | Board | # Played Games | White | Draw | Black | Average points for White |
|---|---|---|---|---|---|---|
| 518 | RNBQKBNR | 20000 | 24.4% | 58.1% | 17.4% | 0.535 |
| 534 | RNBKQBNR | 20000 | 31.7% | 51.4% | 16.9% | 0.574 |
Results for all 960 starting positions
Please send an email to [email protected] and to [email protected] for any feedback.
-
Do you have comments about our Stockfish setup?
- Do you see better ways to generate datasets?
- Would it be reasonable to increase the database size?
-
Are there reasonable ways to group openings into categories?
- The board setup might suggest certain types of opening strategies.
- Which properties of the board configuration imply which types of openings?
The following two projects have both analyzed Chess960 games from Lichess.
An analysis of more than 4 million Chess960 games from Lichess has been conducted here. We represent their data in our format for comparison.
They conclude that "white pieces have an advantage, [and that] the positions setup where black have an advantage are expressively less that positions where white won more."
Using A/B testing, 14 million Chess960 games from Lichess were analyzed here.
They conclude that "there are no starting positions that favor any of the players more than other positions."
- Galen Dorpalen-Barry (Texas A&M, USA)
- Christian Stump (Ruhr University Bochum, Germany)
- The experiments were conducted using large computers at both involved universities.
- The authors thank Nathan Chapelier-Laget, Torsten Hoge, and Alexander Ivanov for useful discussions.
The work in this repository is licensed under the CC BY-NC license. The license is found here.