tmwr-ch7-workflows-classification.Rmd

---
title: "Tidy Modeling with R: Chapter 7 (Workflow Classification)"
output:
  html_document: default
  pdf_document: default
---

```{r performance-setup, include = FALSE}
library(tidymodels)
library(tidyverse)
tidymodels_prefer()
```

This notebook works through the contents of chapter 7 of the book TMWR on model workflows. Workflows encourage good methodology since they provide a single point of entry to the estimation components of a modeling process.  Second, it enables the user to improve project organization.

The data used is the Watson job attrition dataset, which is available from the `modeldata` package.
```{r data}
data(attrition)
glimpse(attrition)
```

We split the data 80/20 into training and test sets, stratified by the outcome variable `Attrition`.
```{r split}
set.seed(502)
attr_split <- initial_split(attrition, prop = 0.80, strata = Attrition)
attr_train <- training(attr_split)
attr_test  <-  testing(attr_split)
```

# Simple Logistic Regression Workflow

A workflow always requires a `parsnip` model object, so we create a logistic regression model with `parsnip`.
```{r}
attr_lm <-
    logistic_reg() %>%
    set_engine("glm") %>%
    set_mode("classification")
```

Let's add the model to a workflow.
```{r}
attr_workflow <- 
    workflow() %>%
    add_model(attr_lm)
attr_workflow
```

We can add formula via the `add_formula()` method.
```{r}
attr_workflow <-
    attr_workflow %>%
    add_formula(Attrition ~ Age + DailyRate)
attr_workflow
```

We can then fit the model through the workflow object.
```{r}
lm_fit <- fit(attr_workflow, attr_train)
tidy(lm_fit)
```

Once a workflow has been fitted, we can use it to make predictions.
```{r}
predict(lm_fit, attr_test %>% slice(1:3))
```

# Evaluating the Test Set

Let's say that we've concluded our model development and have settled on a final model. There is a convenience function called `last_fit()` that will fit the model to the entire training set and evaluate it with the testing set.
```{r}
final_lm_res <- last_fit(attr_workflow, attr_split)
final_lm_res
```

The `.workflow` column contains the fitted workflow and can be pulled out of the results using:

```{r}
fitted_lm_wflow <- extract_workflow(final_lm_res)
```

Similarly, `collect_metrics()` and `collect_predictions()` provide access to the performance metrics and predictions, respectively.

```{r}
collect_metrics(final_lm_res)
```

```{r}
collect_predictions(final_lm_res) %>% slice(1:5)
```

# References

  1. Max Kuhn and Julia Silge, _Tidy Modeling with R_, chapter 7. https://www.tmwr.org/workflows.html (2022)