Love for Linear Models

2b651c3725763904a603ab0a63a46cc8?s=47 Alex Gold
September 17, 2019

Love for Linear Models

A lightning talk extolling the virtues of simple linear models, presented at Data Science DC meetup.

2b651c3725763904a603ab0a63a46cc8?s=128

Alex Gold

September 17, 2019
Tweet

Transcript

  1. LINEAR MODELS ARE GREAT ALEX GOLD SOLUTIONS ENGINEER RSTUDIO @alexkgold

    Slides at: https://github.com/akgold/dsdc_linear_models
  2. IT ME.

  3. WHAT DOES IT MEAN TO BE LINEAR? + some prediction

    error +βhp = β0 + βcyl Just a number
  4. WHAT DOES IT MEAN TO BE LINEAR? Y = βX

    + ϵ
  5. ALEX <3 LINEAR MODELS

  6. None
  7. Y = βX + ϵ

  8. Y = ∑ k βk fk (xi ) + ϵ

  9. None
  10. None
  11. None
  12. gender = β0 + β1 weight + β2 height

  13. gender = β0 + β1 weight + β2 height VS

  14. gender = β0 + β1 weight + β2 height VS

  15. None
  16. It’s all about the Data-Generating Process

  17. mpg = β0 + β1 cyl + disp mpg =

    β0 + β1 cyl + β2 cyl2 + β3 log(disp) OR ?
  18. None
  19. 1. INTERPRETATION MATTERS. 2. LINEARITY ISN’T RESTRICTIVE. 3. MO’ SQUIGGLY

    = MO’ OVERFITTING. 4. SMALL DATA’S OK. 5. IT’S ALL ABOUT THE DGP. @alexkgold https://github.com/akgold/dsdc_linear_models