metadata
license: mit
tags:
- reinforcement-learning
- card-games
- estimation
RL v2 (Yosif's trained bot)
By Yosif Elessawi, published with his permission under the MIT licence. Bid and card networks, copied from a look-ahead bot at the start and then trained by self-play, with 816,000 games in the final stage. No human games were used. It is the benchmark the other bots are measured against. final.pt is a PyTorch checkpoint holding only weights and numbers. rlv2.npz holds the same weights as NumPy arrays, for the GPU self-play code in jaxenv/rlv2.py.
Rated on the Estimation Arena leaderboard. Code and how to run it: estimation-arena. Running it currently needs the rules engine, which is private.