In ryan-heslin/RegLesson: Linear Regression from Scratch

knitr::opts_chunk$set(
    collapse = TRUE,
    comment = "#>",
    fig.path = "man/figures/README-",
    out.width = "100%"
)

RegLesson

The summer before the second year of my master's degree, I identified two weaknesses in my skills: I didn't know enough about linear regression, and I had grown a bit too dependent on tidyverse over base R. The solution was obvious: write a package implementing linear regression from the ground up, using only base R. Instead of relying on lm, I built a LinRegProblem class to store a linear model and associated data. I extended this with functions designed to make it easier to solve textbook problems. I don't need a profiler to know this approach is a lot less efficient than R's built-in functions, but implementing it was a useful exercise.

You can create LinRegProblem objects like so:

library(RegLesson)

X <- LinRegProblem(mtcars[, -1], mtcars$mpg)

This represents a regression of mpg on all other mtcars variables.

the do_problems function evaluates expressions in an environment and prints the results. Required, omitted arguments are replaced with their values in the environment. Now we can check whether the model passes the overall $F$ test and get Working-Hoteling confidence bands for the observations.

do_problems(
    X,
    regression_relation(),
    Working_Hoteling(X_h = cbind(1, as.matrix(mtcars[, -1])))
)

This package contains functions for several diagnostic tests as well. Have fun!

Installation

You can install the development version of RegLesson from GitHub with:

# install.packages("devtools")
devtools::install_github("ryan-heslin/RegLesson")

ryan-heslin/RegLesson documentation built on Aug. 5, 2022, 9:03 p.m.