Skip to content
Distributed proximal LBFGS
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

DPLBFGS - A Distributed Proximal LBFGS Method for Regularized Optimization

This code implements the algorithm used in the experiment of the following paper in C/C++ and MPI: Lee, Ching-pei, Lim, Cong Han, Wright, Stephen J. A Distributed Quasi-Newton Algorithm for Empirical Risk Minimization with Nonsmooth Regularization. The 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2018.

In additional to our algorithm, the following algorithms are also implemented.


Getting started

To compile the code, you will need to install g++ and an implementation of MPI. You will need to list the machines being used in a separate file, and make sure they are directly accessible through ssh. Additionally the code depends on the BLAS and LAPACK libraries.

The code, borrowed from MPI-LIBLINEAR, partition the data and distribted the segments to the designated machines. Then the program ./train solves the optimization problem to obtain a model.

Problem being solved

The code solves the L1-regularized logistic regression problem.

min_{w} |w|1 + C \sum{i=1}^n \log(1 + \exp(- y_i w^T x_i))

with a user-specified parameter C > 0.

You can’t perform that action at this time.