13
0

Best LpL_p Isotonic Regressions, p{0,1,}p \in \{0, 1, \infty\}

Abstract

Given a real-valued weighted function ff on a finite dag, the LpL_p isotonic regression of ff, p[0,]p \in [0,\infty], is unique except when p[0,1]{}p \in [0,1] \cup \{\infty\}. We are interested in determining a ``best'' isotonic regression for p{0,1,}p \in \{0, 1, \infty\}, where by best we mean a regression satisfying stronger properties than merely having minimal norm. One approach is to use strict LpL_p regression, which is the limit of the best LqL_q approximation as qq approaches pp, and another is lex regression, which is based on lexical ordering of regression errors. For LL_\infty the strict and lex regressions are unique and the same. For L1L_1, strict q1q \scriptstyle\searrow 1 is unique, but we show that q1q \scriptstyle\nearrow 1 may not be, and even when it is unique the two limits may not be the same. For L0L_0, in general neither of the strict and lex regressions are unique, nor do they always have the same set of optimal regressions, but by expanding the objectives of LpL_p optimization to p<0p < 0 we show p0p{ \scriptstyle \nearrow} 0 is the same as lex regression. We also give algorithms for computing the best LpL_p isotonic regression in certain situations.

View on arXiv
Comments on this paper