Radar ·
RL replay buffer with top-k in SQL
A post dated August 16, 2026 describes weighted top-k SQL sampling for a reinforcement-learning replay buffer and reports selecting 8,192 rows from 50 million in about 1.1 seconds on a laptop.
Published on August 16, 2026, the post presents Replayhouse, which uses a table as a reinforcement-learning replay buffer. It implements Efraimidis–Spirakis weighted top-k sampling in SQL, with samples filterable using WHERE. The post reports selecting 8,192 rows from 50 million in about 1.1 seconds on a laptop.
The stated rationale is that a table-based buffer and SQL filters can simplify replay-sampling workflows. The timing is the post’s reported result, not an independent verification or a guarantee for other environments. To assess it, consult the original post and check its method, query, and test conditions; compare with a reproduction using equivalent data and infrastructure.
Original source: Replay buffer de RL com amostragem ponderada top-k em SQL