Guess model for Self-taught + guesses
For each card a player cannot see, guesses which of the other three players holds it. It reads the same view of the table as the Self-taught model and was trained only on the Self-taught model's games against itself, with no human games and nothing from other bots. The file is a NumPy .npz of the network's weights (three hidden layers of 512).
The Self-taught + guesses bot uses it to imagine the hidden hands it searches over.
Rated on the Estimation Arena leaderboard. Code and how to run it: estimation-arena (train/guess_net.py, bots/guess-search). Running it currently needs the rules engine, which is private.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support