grad_descent: Gradient Descent

In brian-d1018/bis557: Brian's BIS 557 Work

Description

This function uses the gradient descent algorithm (matrix form) to solve the coefficients of simple linear regression with least squares error. This is an optimization algorithm to minimize a function which updates the parameters of the model. A very common statistical and ML algorithm.

Usage

grad_descent(X, y, b_0, learn_rate, max_iter)

Arguments

X	matrix of all the predictors (excludes the column of 1's)
y	column vector of target (response) values
b_0	parameters of model: column vector of initialized coefficients
learn_rate	the initialized learning rate (aka step size)
max_iter	the maximum number of iterations for this algorithm

Value

the estimated coefficients/parameters

Examples

data(iris)
print(b <- grad_descent(X = iris[,2:4], y = iris[,1], b_0 = rep(1e-16, 4),
                        learn_rate = 2, max_iter = 2e5))

brian-d1018/bis557 documentation built on Dec. 17, 2020, 6:21 p.m.