关键词: 近似查询处理, 自回归模型, 数据库, 深度学习, 聚合函数

Abstract: Abstract: Approximate Query Processing (AQP) of aggregate functions was recently a research hotspot in the database field. Existing approximate query techniques have problems such as high query response time, high storage overhead, and no support for multi-predicate queries. Thus, a Deep autoregressive-based Approximate Query Processing method (DeepAQP) was proposed. DeepAQP learned the joint probability distribution of multi-column data in the table using a deep autoregressive model and estimated the target column's selectivity and distribution, which is limited by the query predicates. Thus, DeepAQP could effectively handle the approximate query requests of aggregation functions with multiple predicates in a single table. Experiments were conducted on the TPC-H and TPC-DS datasets. Compared with Verdict DB, which is a sample-based method, the query response time was reduced by 2 to 3 orders of magnitude, and the storage space was reduced by 3 orders of magnitude. Compared with DBEst++, which is a machine learning-based method, the query response speed was also reduced by 1 order of magnitude, and the model training time was reduced significantly. Besides, DeepAQP could handle multi-predicate query requests, which DBEst++ does not support. Experimental results show that DeepAQP achieves a better trade-off between accuracy and speed while reducing the training and storage overhead significantly.

Key words: approximate query processing, autoregressive model, database, deep learning, aggregate functions