Conditional Generative Model Based Predicate-Aware Query Approximation

The goal of Approximate Query Processing (AQP) is to provide very fast but "accurate enough" results for costly aggregate queries thereby improving user experience in interactive exploration large datasets. Recently proposed Machine-Learning based AQP techniques can low latency as query execution only involves model inference compared traditional processing on database clusters. However, with increase the number filtering predicates(WHERE clauses), approximation error significantly increases these methods. Analysts often use a predicates insights discovery. Thus, maintaining important prevent analysts from drawing misleading conclusions. In this paper, we propose ELECTRA, predicate-aware system that answer analytics-style much smaller errors. ELECTRA uses conditional generative learns distribution data and at runtime generates small (~1000 rows) representative sample, which executed compute approximate result. Our evaluations four different baselines three real-world datasets show provides lower baselines.