Research project
Mastering Leduc Hold’em
Comparing reinforcement learning and counterfactual regret minimization in imperfect-information poker.
2024Repository
Research overview
We trained 16 reinforcement learning agents in RLCard and compared them with counterfactual regret minimization and an aggressive RaiseCall baseline. CFR+ ranked highest in the reported Elo and big-blind evaluations. The full report gives the setup and results.