So went from we use our own data, not stockfish’s data. To it’s a community binpack. Which stockfish filtered and uses for their nets. So you have done nothing different with the nnue.FireDragon761138 wrote: ↑Sun Mar 29, 2026 12:05 pmNo, we trained our own NNUE using a community binpack. So the NNUE is our own.
The code has also had substantial modifications, contrary to what people are saying, Theoria is not merely a clone. We are still working on it, but development has slowed as it's functional. In the future we'll try to make refinements to improve raw Elo performance without losing interpretability. We've had several positive comments from testers so far.
Sorry about the delay in the response. I've mostly moved on to other projects and activities. We are currently developing a website that uses Theoria to evaluate games as a web application, but it's mostly what my partner is doing, not myself. I'm hoping eventually to get somebody else interested in the ideas behind the project, so it's mostly meant to be a functional proof of concept at this point.
Theoria 0.2 is out
Moderator: Ras
-
lucario6607
- Posts: 53
- Joined: Sun May 19, 2024 5:44 am
- Full name: Kolby Mcgowan