The tidymodels ecosystem

Modeling with tidymodels in R

David Svancer

Data Scientist

Collection of machine learning packages

The tidymodels package

Collection of machine learning packages

Data resampling with rsample

Collection of machine learning packages

Feature engineering with recipes

Collection of machine learning packages

Model fitting with parnsip

Collection of machine learning packages

Model tuning with tune and dials

Collection of machine learning packages

Model evaluation with yardstick

Supervised machine learning

Branch of machine learning that uses labeled data for model fitting

Regression

Predicting quantitative outcomes
- Selling price of a home

Classification

Predicting categorical outcomes
- Whether an employee will leave a company

left_company	miles_from_home	salary
no	1	84500
yes	10	64820
no	5	76490
yes	19	68540

tidymodels variable roles

left_company is an outcome variable
miles_from_home and salary are predictor variables

Data resampling

Create training and test sets

Guards against overfitting
Common ratio is 75% training, 25% test

Training data

Feature engineering
Model fitting and tuning

Test data

Estimate model performance on new data

Creating training and test datasets

Fuel efficiency data

Vehicle fuel efficiency data from the U.S. Environmental Protection Agency

Outcome variable is hwy - highway fuel efficiency in miles per gallon (mpg)

mpg

# A tibble: 234 x 11
     hwy   cty displ   cyl manufacturer model       year trans      drv   fl    class  
   <int> <int> <dbl> <int> <chr>        <chr>      <int> <chr>      <chr> <chr> <chr>  
 1    29    18   1.8     4 audi         a4          1999 auto(l5)   f     p     compact
 2    29    21   1.8     4 audi         a4          1999 manual(m5) f     p     compact
 3    31    20   2       4 audi         a4          2008 manual(m6) f     p     compact
 4    30    21   2       4 audi         a4          2008 auto(av)   f     p     compact
 5    26    16   2.8     6 audi         a4          1999 auto(l5)   f     p     compact
# ... with 224 more rows

Data resampling with tidymodels

initial_split()
- Specifies instructions for creating training and test datasets
- prop specifies the proportion to place into training
- strata provides stratification by the outcome variable
Pass split object to training() function

Pass split object to testing() function

library(tidymodels)

mpg_split <- initial_split(mpg,
                           prop = 0.75,
                           strata = hwy)

mpg_training <- mpg_split %>%
  training()

mpg_test <- mpg_split %>%
  testing()

Home sales data

Home sales from the Seattle, Washington area between 2015 and 2016

home_sales

# A tibble: 1,492 x 8
   selling_price home_age bedrooms bathrooms sqft_living sqft_lot sqft_basement floors
           <dbl>    <dbl>    <dbl>     <dbl>       <dbl>    <dbl>         <dbl>  <dbl>
 1        487000       10        4      2.5         2540     5001             0      2
 2        465000       10        3      2.25        1530     1245           480      2
 3        411000       18        2      2           1130     1148           330      2
 4        635000        4        3      2.5         3350     4007           800      2
 5        380000       24        5      2.5         2130     8428             0      2
# ... with 1,482 more rows

Let's practice!

Modeling with tidymodels in R