Printable Version

Back to Previous Page

Stochastic Gradient Perceptron

This command develops a linear discriminant using a stochastic gradient method. Formally speaking, a one-layer network with one perceptron is used. The user can set three parameters: maximum iterations, learning rate and learning rate decay. The larger the learning rate parameter, the larger the weights vector changes in each iteration. The method makes use of the sigmoid function as an approximation of the binary response of a two-class problem.