glm_grad: A first-order solution for the GLM maximum likelihood problem

In wuyinfeistella/bis557: Linear model and the gradient descent

Description

a first-order solution for the GLM maximum likelihood problem using only gradient information without the Hessian matrix

Usage

glm_grad(
  X,
  y,
  family = poisson(link = "log"),
  niter = 1000,
  lambda = 1e-05,
  update = FALSE,
  gamma = 0.95
)

Arguments

X	The predictors of the glm
y	The response variable of the glm
family	The link function
niter	The number of iterations
lambda	The learning rate
update	Either a constant step size or an adaptive step size
gamma	The momentum step size

Value

The estimated coefficients for the glm

Examples

#reference CASL 5.3
n = 1000; p = 5
X <- cbind(1, matrix(rnorm(n*(p-1)), ncol=p-1))
y = sample(c(0,1), replace=TRUE, size=n)
glm(X, Y, update=FALSE)

wuyinfeistella/bis557 documentation built on Jan. 1, 2021, 12:52 p.m.