In joepowers16/rethinking: Statistical Rethinking book package

Overview:

Adding parameters will always improve a model's fit as measured by R^2^.

knitr::opts_chunk$set(
  fig.align='center', 
  #dpi = 150, 
  include=FALSE, echo=FALSE, message=FALSE, warning=FALSE
)

library(magrittr)
library(modelr)
library(tidyverse)

file_r2_explained <- paste(dir_images, "r2_explained.png", sep = "/")
file_r2_squared_formula <- paste(dir_images, "r_squared_formula.png", sep = "/")

knitr::include_graphics(file_r2_explained)
# image from kindle p167

knitr::include_graphics(file_r_squared_formula)
# image from http://www.simages.org/r-squared-formula/

The underfitting-overfitting problem is often described as the bias-variance trade-off.

Takeawys

Adding parameters will always improve a model's fit as measured by R^2^, but it will produce overfit models that predict poorly. Underfit models will be insensitive to their data. These pressures need to be balanced to build the most informative models.

joepowers16/rethinking documentation built on June 2, 2019, 6:52 p.m.