Iris

This page contains the abbreviated code for the iris modeling example. For a detailed walkthrough, look at the RSample through Workflows tutorials. Note that this example uses a workflow object to create the final predictions, and thus the recipe and parsnip model are not prepped or fit independently.

library(tidymodels)

# defining data_iris
data_iris <- bind_cols(all_of(iris), 
                       row = 1:nrow(iris))

# RSample - generate data splits 
data_split <- initial_split(data_iris, prop = 3/4)

testing <- testing(data_split)
training <- training(data_split)

# Recipes - define and apply preprocessing steps
iris_recipe <- recipe(Species ~ ., 
                      data = training) |>
  update_role(row, new_role = "ID") |>
  step_corr(all_numeric_predictors()) |>
  step_normalize(all_numeric_predictors()) |>
  step_rename(Row = row) 

# Parsnip - create a model
rf <- rand_forest() |>
  set_mode("classification") |>
  set_args(trees = 200) |>
  set_engine("randomForest")

# Workflows - bundle recipe and model together and generate predictions
iris_wkf <- workflow() |>
  add_recipe(iris_recipe) |>
  add_model(rf) |>
  fit(training)

iris_preds <- iris_wkf |>
  augment(testing)

# Yardstick - evaluate performance 
my_metrics <- metric_set(sens, spec, accuracy)
iris_preds |>
  my_metrics(truth = Species, estimate = .pred_class)

## # A tibble: 3 × 3
##   .metric  .estimator .estimate
##   <chr>    <chr>          <dbl>
## 1 sens     macro              1
## 2 spec     macro              1
## 3 accuracy multiclass         1

iris_preds |>
  roc_curve(Species, .pred_setosa:.pred_virginica) |>
  autoplot()