Reinforcement Learning Lab

Train a tabular Q-learning policy through self-play against a rotating heuristic opponent pool. Completed policies are stored as activatable models.

Episodes i Steps / Episode i Alpha i Gamma i Start RL Job

Exploration schedule is fixed in this UI: epsilon starts at 1.0, decays by 0.995 each episode, and floors at 0.05.

How the opponent curriculum works

Default schedule is curriculum: early episodes focus on basics, middle episodes emphasize adaptive heuristics, and late episodes emphasize tougher opponents. Default pool: rock, paper, scissors, copy_opponent, reactionary, counter_reactionary, statistical, markov, nash_equilibrium, multi_armed_bandit.

RL Job Status

No active RL job.

Recent RL Jobs

Refresh RL Jobs

ID	Status	Progress	Model ID	Final 50 Win Rate