gradient.descent: Gradient descent method to find the coefficients of linear...

In Lrrzz/StatComp21064: Example functions for 'Statistical Computing' course

Description

Gradient descent method to find the coefficients of linear regression equation

Usage

gradient.descent(
  x,
  y,
  epsilon = 1e-14,
  maxit = 1000,
  stepsize = 0.001,
  alpha = 0.25,
  b = 0.8,
  stepmethod = "backtracking",
  verbose = TRUE,
  plotLoss = TRUE
)

Arguments

x	a m*n data matrix
y	observations size of m*1
epsilon	termination condition, difference between the coefs of two consecutive iters
maxit	the maximum number of iterations, default is 1000
stepsize	just stepsize, must smaller than 1e-3, default is 1e-3
alpha	parameter in the backtracking method, should in (0,0.5), default is 0.25
b	parameter in the backtracking method, should in (0,1), default is 0.8
stepmethod	optional "backtracking" or "fixed", default is "backtracking"
verbose	whether to print out iterations, default is "TRUE"
plotLoss	whether to plot loss, default is "TRUE"

Value

a list contains coefficients,RSE and number of iterations

Examples

## Not run: 
n=100
x1<-rnorm(n)
x2<-rnorm(n)
y=1+0.5*x1+x2+rnorm(n,sd=0.1)
x<-cbind(x1,x2)
gradient.descent(x,y)

## End(Not run)

Lrrzz/StatComp21064 documentation built on Dec. 23, 2021, 10:21 p.m.