Complexity Analysis of the Lasso Regularization Path
Abstract
The regularization path of the Lasso can be shown to be piecewise linear, making it possible to "follow" and explicitly compute the entire path. We analyze in this paper this popular strategy, and prove that its worst case complexity is exponential in the number of variables. We then oppose this pessimistic result to an (optimistic) approximate analysis: We show that an approximate path with at most O(1/sqrt(epsilon)) linear segments can always be obtained, where every point on the path is guaranteed to be optimal up to a relative epsilon-duality gap. We complete our theoretical analysis with a practical algorithm to compute these approximate paths.
Cite
@article{arxiv.1205.0079,
title = {Complexity Analysis of the Lasso Regularization Path},
author = {Julien Mairal and Bin Yu},
journal= {arXiv preprint arXiv:1205.0079},
year = {2012}
}
Comments
To appear in the proceedings of 29th International Conference on Machine Learning (ICML 2012)