sgd: Stochastic Gradient Descent

In tbonza/supml: "Algorithms for Supervised Machine Learning"

Description

Stochastic Gradient Descent

Usage

sgd(alpha, X, y, grad, N, precision = 1e-04)

Arguments

alpha	learning rate constant
X	matrix of predictors including 1 for beta 0
y	vector of dependent variable
grad	gradient function
N	maximum number of iterations
precision	constant used for approximation condition

Value

list object with theta, convergence status, and number of iterations

Examples

x <- runif(1000, -5, 5)
y <- x + rnorm(1000) + 3
X <- cbind(1, matrix(x))
h <- sgd(0.1, X, y, least_squares_gradient, 1000)
h$theta
h$converged == TRUE
h$iterations

tbonza/supml documentation built on May 17, 2019, 3:14 a.m.