Minimizes function using Stochastic Gradient Descent Algorithm. Variation of the L. Bottou's SGD and Inria's JSGD. This version allows to use arbitrary objective function via the following interface (similar to Schmidt's minFunc): sgd(funObj, funPrediction, x0, train, valid, options, varargin)
I provide the source code together with the example (softmax objective function).
gd_matlab is a Gradient Descent method similar to SGD. Here the idea is that instead of using SGD we use just simple GD and delegate the responsibility of computing (noisy) gradient to the objective function.