For some reason, when slot machines went digital, they lost their best feature — the handle. Who wants to push a button on a slot machine, anyway? Might as well just play video poker. [John Bradnam] ...

The canonical 3×3 Gauss–Jordan plan takes 9 steps on a full matrix (6 eliminations + 3 normalisations) and 6 steps on a triangular matrix (3 eliminations above the diagonal + 3 normalisations). Step ...

This is the 3×3 extension of rref_rl/. It applies the same Double DQN + sparse-reward + action-masking recipe, but uses a two-phase curriculum to learn the larger 12-action space: Phase 1 — train on ...