We are happy to announce the first release of the Open Chess Plies, a high-quality dataset with evaluated positions.
What Is OCP?
OCP is a chess dataset that relies on strong, open-source chess engines to play different tournaments against each other and collect the positions.
It focuses on quality, in lieu of quantity, which makes it different than most other chess datasets.
The Philosophy
We're not just about quality; we're also about diversity. We use a lot of samples from the UHO, as well as the Lichess elite database.
The OCP also includes FRC and DFRC (Double FRC), which is a must-have in modern Chess ML.
It is perfect for mixing with other datasets, giving the OCP dataset more weight.
Two Releases Every Month
Every month, we release all evaluated positions from all the games.
Somewhat later in the month we release a cleaned dataset that is even better, with the downside being, that it is smaller.
Why You Should Care
If you are into ML and DL, you know that you can't immediately see what's behind a Stockfish or Lc0 binpack. That's why we built OCP, a fully open dataset for everyone to use as they wish.
Getting Started
You can find all releases here. Download the JSONL file and convert it to binpack, bullet-format, or whatever you want.
