流数据下的复合分位数回归
Renewable Composite Quantile Regression for Streaming Data Sets
韩星敏 1姜荣2
作者信息
- 1. 东华大学 理学院,上海 201600
- 2. 上海第二工业大学 数理与统计学院,上海 201209
- 折叠
摘要
随着互联网的发展,数据规模急剧增长,但有限的内存只能存储一小批数据,因此在不访问历史数据的情况下进行分析是非常有必要的,流数据分析也因而引起了广泛关注.同时复合分位数回归因其鲁棒性和全面性,在许多领域得到应用,但由于传统复合分位数回归是基于内存可容纳完整数据的条件,因此在流数据环境中实现复合分位数回归是非常有挑战的.针对流数据提出了一种可更新的复合分位数回归方法,可以随着数据的到达,使用当前数据和历史数据的汇总统计量来更新估计量.在理论上证明提出的可更新估计量与使用完整数据得到的估计量是渐近等价的.最后通过模拟研究验证了所提出方法的有效性.
Abstract
With the development of the Internet,the scale of data has grown dramatically.However,due to limited memory capacity that can only store a small batch of data,it is essential to analyze data without accessing historical data.Consequently,streaming data analysis has attracted widespread attention.Meanwhile,composite quantile regression,known for its robustness and and comprehensiveness,has been applied in various fields.However,implementing composite quantile regression for streaming data is challenging since traditional methods are based on the condition that the entire dataset can fit into memory.An updating composite quantile regression method specifically is designed for streaming data.The estimates can be updated as the data arrives using both current data and the summary statistics of historical data.In theory,it is proven that the updatable estimator proposed is asymptotically equivalent to the estimator obtained using complete data.Finally,the effectiveness of the proposed method is verified through simulation research.
关键词
复合分位数回归/流数据/在线可更新估计方程Key words
composite quantile regression/streaming data/online updating estimating equation引用本文复制引用
出版年
2024
NETL
NSTL
万方数据