Created as a part of EECS 545 (Machine Learning) course at University of Michigan
Data : Download the file nuclear.mat. The variables x and y contain training data for a binary classification problem.
The variables correspond to the total energy and tail energy of waveforms producedby a nuclear particle detector.
The classes correspond to neutrons and gamma rays.
Implementation :
L is hinge loss
We first implement subgradient method for minimizing J and apply it to the nuclear data and then implement stochiastic subgradient method, which is like the subgradient method, except that the step direction is a subgradient of a random Ji, not J.