Sampling techniques for supervised and unsupervised tasks

Abstract

Survey sampling considers selecting a random sample of units to represent a larger population. For decades, national statistical agencies have worked with samples of reasonably large size, aiming at shortening the delay of production of statistical information, with limited cost. For the past 10 years, we have been facing a completely new situation with a burst of the volume of the generated data. Since these data cannot be saved and treated comprehensively, sampling is needed to select large but tractable samples, sometimes processed in real time, to faithfully represent these huge masses of data. More than ever, we need to control the statistical properties of the estimators produced. The purpose of this chapter is therefore to provide an introduction to the estimation framework of survey sampling, by reviewing some of the classical sampling methods.