Surprise sampling: Improving and extending the local case-control sampling

Fithian and Hastie (2014) proposed a new sampling scheme called local case-control (LCC) that achieves stability efficiency by utilizing clever adjustment pertained to the logistic model. It is particularly useful for classification with large imbalanced data. This paper proposes more general based on working principle data points deserve higher probability if they contain information or appear "surprising" in sense of, example, error of pilot prediction absolute score. Compared relevant existing schemes, as reported Ai, et al. (2018), one has several advantages. adaptively gives out optimal forms variety objectives, including LCC Ai (2018)'s special cases. Under same model specifications, estimator also performs no worse than those literature. The estimation procedure valid even misspecified and/or inconsistent dependent full We present theoretical justifications claimed advantages optimality design. Different from our sample theory are population-wise rather data-wise. Moreover, approach can be applied unsupervised learning studies, since it essentially only requires specific loss function response-covariate structure needed. Numerical studies carried evidence support shown.