Semi Stochastic Gradient Descent 1.0

jakub konecny — Wed, 09 Jul 2014 04:28:47 -0000

Efficient implementation of Semi-Stochastic Gradient Descent algorithm (S2GD) for training logistic regression (L2-regularized).

The S2GD algorithm enjoys linear convergence (faster than SGD), and comes without the need of tuning parameters like stepsize (choice explained in paper).

The code is in C++, called from MATLAB. The code is well commented, and should be easily adjusted for different applications.

The package also contains implementation of SGD and SAG.

First time usage:

>> mexAll % compiles the .cpp files

>> demo

mloss.org Semi Stochastic Gradient Descent

Semi Stochastic Gradient Descent 1.0