Codes for subsampled Newton methods for solving ridge logistic regression.
Consider minimizing a sum of convex functions. Subsampled-Newton methods subsample the functions to calcuate the approximated Hessian. Using non-uniform sampling schemes, we can show the sampling size can be independent of number of functions. In typical ERM problems where n >> d, sub-sampled Newton methods can speed up a lot.
-
Matlab users: See
subsampled_newton.mfor main functions. And seedemo_comparison.mfor usage. -
Python users: See files in the folder
python.
Peng Xu, Jiyan Yang, Farbod Roosta-Khorasani, Christopher Ré, and Michael W. Mahoney, Sub-sampled Newton Methods with Non-uniform Sampling, NIPS 2016.