Aggregate-based Training Phase for ML-based Cardinality Estimation

Abstract Cardinality estimation is a fundamental task in database query processing and optimization. As shown recent papers, machine learning (ML)-based approaches may deliver more accurate cardinality estimations than traditional approaches. However, lot of training queries have to be executed during the model phase learn data-dependent ML making it very time-consuming. Many those or example use same base data, structure, only differ their selective predicates. To speed up phase, our core idea determine predicate-independent pre-aggregation data execute over this pre-aggregated data. Based on idea, we present specific aggregate-based for ML-based paper. are going show with different workloads evaluation, able achieve an average speedup 90 thus outperform indexes.