04_nested_dataframes.Rmd

---
title: "Nesting data frames"
author: "John Little"
date: "`r Sys.Date()`"
output: html_notebook
---

One way to leverage what we've learned in previous lessons (functions, purrr/map, regex, ggplot2, dplyr) is to use the `nest` function.  Nesting will create group-specific tibbles upon which which you can map more functional analysis.

```{r}
library(tidyverse)
```

## `group_by` and nest

While `group_by` is not specifically required (see next code-chunk) to `nest`, I find that it helps me when I'm trying to leverage ggplot functions as you'll see below.

```{r}
starwars |>
  group_by(gender) |>
  nest()
```

### `nest` without group_by


```{r}
starwars |>
  nest(-gender)
```

## map a custom function over a nested group

Building on lesson three where we created custom functions, we can map those functions onto our embedded tibbles

```{r}
make_me_a_plot <- function(df) {
  df %>% 
    ggplot(aes(height, mass)) +
    geom_point() 
}

starwars |>
  nest(data = -gender) |>
  mutate(my_plot = map(data, make_me_a_plot)) |>
  pull(my_plot)
```

Adding titles to the plots requires the `group_by()` function

```{r, warning=TRUE}
starwars |> 
  group_by(gender) |>
  nest(my_df = -gender) |> 
  mutate(my_plot = map(my_df, make_me_a_plot)) |> 
  mutate(my_plot_with_title = map(my_plot, ~ .x + ggtitle(gender))) |>
  pull(my_plot_with_title)                     # This can be executed more elegantly with map2(). See Below.
```

### Models and mapping other functions

As with other functions, you can map linear regressions -- `lm()` -- and other functions.  Once the model is embedded as a list within a data frame, you can map the `broom::tidy()` function onto the model.  Then `unnest()` the tidied _model_  and manipulate the data with standard dplyr verbs.

```{r}
starwars |>
  group_by(sex) |>
  nest(-sex) |>
  mutate(my_plot = map(data, make_me_a_plot)) |>
  mutate(my_plott = map(my_plot, ~ .x + labs(title = sex))) |> 
  mutate(my_model = map(data, ~ lm(mass ~ height, data = .x))) |>
  mutate(my_tidy_model = map(my_model, broom::tidy)) #|> 
  # unnest(my_tidy_model) |>
  # filter(term != "(Intercept)") |>
  # filter(p.value < 0.05) |>
  # mutate(my_reg_plot = map(data, ~ .x |>
  #                            ggplot(aes(height, mass)) +
  #                            geom_smooth(method = lm, se = FALSE) +
  #                            geom_point())) |>
  # pull(my_reg_plot)
```


## `map2()` - Iterating over more than a single parallel vector

In the section about mapping custom functions, I use the `group_by()`, `nest()` and `map()` functions to generate plots with titles.  the `.x` pronoun refers to the object I want to iterate over.  That code works fine even if it's a little verbose.  It also demonstrates the value of additional {purrr} functions.  `map2()` iterates over two vectors in parallel. Similarly, `pmap()` iterates over a list of multiple vectors.  

In the case of `map2()` the two vectors of a parent data-frame can be referred to by the pronouns `.x` and `.y`.  So, one vector is the title of the plot (`.y`), the other vector (`.x`) is the list-column tibble that contains the x-axis and y-axis vectors generating a scatter plot for each nested tibble.

Essentially, coding with `map2()` creates the scatter plots with a bit less code

```{r}
starwars |> 
  nest(my_df = -gender) |> 
  mutate(my_plot = map2(my_df, gender, ~ .x |> 
                          ggplot(aes(height, mass)) +
                          geom_point() + 
                          ggtitle(.y))) |>
  pull(my_plot)  
```

Of course, the anonymous function can be written as a named function, making the code even more readable

```{r}
splot_title <- function(my_nested_df, my_title) {
  my_nested_df |> 
    ggplot(aes(height, mass)) +
    geom_point() +
    ggtitle(my_title)
}

starwars |> 
  nest(my_df = -gender) |> 
  mutate(my_plot = map2(my_df, gender, splot_title)) |>
  pull(my_plot)  
```