Weighted Random sampling based hierarchical amnesic synopses for data streams

Abstract

Maintaining a synopsis structure dynamically from data stream is vital for a variety of streaming data applications, such as approximate query or data mining. In many cases, the significance of data item in streams decays with age: this item perhaps conveys critical information first, but, as time goes by, it gets less and less important until it eventually becomes useless. This characteristic is termed amnesic. Random Sampling is often used in construction of synopsis for streaming data. This paper proposed a Weighted Random Sampling based Hierarchical Amnesic Synopses which includes the amnesic characteristic of data stream in the generation of synopsis. The construction methods for weighted random sampling with and without replacement are discussed. We experimentally evaluate the proposed synopsis structure.

Data Summarization Techniques for Big Data—A Survey

“…For example, WRS can be applied over data stream that is considered as big data, to take a sample from the recent data streams since based on weighting different part of data streams, it is possible to choose a part of data streams which has high weight according to recent data streams. There are various extensions of WRS in the literature such as [70] and [71].…”

Recent advances in scaling‐down sampling methods in machine learning

“…In situations where N is not known, such as streaming data or very large data sets, there are several variations of simple random sampling such as Random Sampling with Reservoir , Biased Reservoir Sampling , Acceptance/Rejection Sampling , Chain Sampling , and Weighted Random Sampling. Simple random sampling includes only Step 1 of Algorithm 1.…”