The active-set constraints for the Lasso path

Let A{\mathcal A} be the active set chosen by the Lasso, and let A0{\mathcal A}_0, A1{\mathcal A}_1, and A10{\mathcal A}_{10} be the sets defined in the source. Let AAA_{\mathcal A}, GA{\mathcal G}_{\mathcal A}, wAw_{\mathcal A}, and c^j\widehat c_j denote the corresponding quantities, with

wA=AAGA11A.w_{\mathcal A}=A_{\mathcal A}{\mathcal G}_{\mathcal A}^{-1}{\bf 1}_{\mathcal A}.

Lasso active-set constraints. The Lasso choice must satisfy A1A{\mathcal A}_1\subseteq {\mathcal A} and AA10{\mathcal A}\subseteq {\mathcal A}_{10}; moreover, wAw_{\mathcal A} cannot have sign(wj)sign(c^j)\operatorname{sign}(w_j)\ne\operatorname{sign}(\widehat c_j) for any jA0j\in{\mathcal A}_0, and, subject to these first three constraints, A{\mathcal A} must minimize AAA_{\mathcal A}.

These are consequences of the preceding lemmas and characterize the constraints used to compare the Lasso and LARS--Lasso paths. The source presents them as consequences rather than as an independently named conjecture; their resolution is therefore not separately established here.

Sources & referencesView supporting material

Primary source

Bradley Efron, Trevor Hastie, Iain Johnstone and Robert Tibshirani, “Least Angle Regression”, arXiv:math/0406456 (2004).

Progress summary

Never refreshed

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Solutions 0

No solutions have been posted yet.