We introduce TS-Insight, a tool for visually analyzing the internal decision-making mechanisms of Thompson Sampling (TS)-based algorithms. While TS is effective in balancing exploration and exploitation strategies in active learning, its probabilistic nature makes debugging and reliability difficult. TS-Insight provides verification, diagnosis, and explainability of exploration/exploitation dynamics through multiple plots that track the posterior distribution, evidence count, and sampling results for each arm, thereby enhancing reliability and enabling effective debugging and deployment in complex binary decision-making scenarios, especially in sensitive areas requiring interpretable decisions.
