glm_momentum: Implement gradient descent GLM, adaptive (momentum)

In kimgannon/bis557: Contains Homework Assignments for BIS 557 2020

Description

This function implements gradient descent GLM using the momentum algorithm

Usage

glm_momentum(X, Y, mu_fun, max_n, lr, gamma = 0.8, tol = 1e-10)

Arguments

X	a model matrix, the X variables as columns
Y	a factor vector, the response variable
mu_fun	a function depending on Xbeta, giving the mean of the link function
max_n	max number of iterations (a positive integer)
lr	the learning rate
gamma	the constant weight of the past update (a positive number)
tol	the difference threshold for which we will exit the algorithm for iterations less than max_n

Examples

n <- 5000; p <- 3
beta <- c(-1, 0.2, 0.1)
X <- cbind(1, matrix(rnorm(n*(p-1)), ncol = p-1))
eta <- X %*% beta
lambda <- exp(eta)
y <- stats::rpois(n, lambda = lambda)
glm_momentum(X, y, mu_fun = function(eta) exp(eta), lr=0.00000001, max_n = 100000)

kimgannon/bis557 documentation built on Nov. 25, 2020, 7:09 a.m.