purrr mapping not producing tidy data - r

Question

Thanks to this site, I'm using the R purrr package to aggregation data based on multiple columns. The aggregation is working how I want but the output is not. Here is a sample using the mtcars dataset.
library(dplyr)
library(purrr)
#pull in data
data <- mtcars
#get colnames
variable1 <- colnames(data)
#map the variables
t1 <- map(variable1, ~ data %>%
group_by_at(.x) %>%
summarize(number = mean(mpg))) %>%
set_names(variable1) %>%
bind_rows(., .id = 'variable')
Were I expect three columns (Predictor Variable, Levels within Each of those Variables, aggregation), I have 8. See the image below:
How can I take my code up at the top and turn out a tidy dataset?

Answer 1

A simple way to do this is to reshape your data to long form, which lets you aggregate with ordinary dplyr:
library(tidyverse)
mpg_means <- mtcars %>%
gather(variable, value, -mpg) %>%
group_by(variable, value) %>%
summarise(mean_mpg = mean(mpg))
mpg_means
#> # A tibble: 146 x 3
#> # Groups: variable [?]
#> variable value mean_mpg
#> <chr> <dbl> <dbl>
#> 1 am 0. 17.1
#> 2 am 1. 24.4
#> 3 carb 1. 25.3
#> 4 carb 2. 22.4
#> 5 carb 3. 16.3
#> 6 carb 4. 15.8
#> 7 carb 6. 19.7
#> 8 carb 8. 15.0
#> 9 cyl 4. 26.7
#> 10 cyl 6. 19.7
#> # ... with 136 more rows
Note that while mtcars is entirely numeric, if you have different types, converting to long form will coerce variable types. The calculations will be the same, but it may cause issues later. To resolve it, use an output format that can handle diverse types, e.g.
mpg_means_in_list_cols <- mtcars %>%
as_tibble() %>% # compact printing for list columns
summarise_all(list) %>% # collapse each column into a list of itself
gather(group, group_values, -mpg) %>%
mutate(mpg_means = map2(mpg, group_values, # for each mpg/value pair, ...
~tibble(mpg = .x, group_value = .y) %>% # ...reconstruct a data frame...
group_by(group_value) %>%
summarise(mean_mpg = mean(mpg)))) # ...and aggregate
mpg_means_in_list_cols
#> # A tibble: 10 x 4
#> mpg group group_values mpg_means
#> <list> <chr> <list> <list>
#> 1 <dbl [32]> cyl <dbl [32]> <tibble [3 × 2]>
#> 2 <dbl [32]> disp <dbl [32]> <tibble [27 × 2]>
#> 3 <dbl [32]> hp <dbl [32]> <tibble [22 × 2]>
#> 4 <dbl [32]> drat <dbl [32]> <tibble [22 × 2]>
#> 5 <dbl [32]> wt <dbl [32]> <tibble [29 × 2]>
#> 6 <dbl [32]> qsec <dbl [32]> <tibble [30 × 2]>
#> 7 <dbl [32]> vs <dbl [32]> <tibble [2 × 2]>
#> 8 <dbl [32]> am <dbl [32]> <tibble [2 × 2]>
#> 9 <dbl [32]> gear <dbl [32]> <tibble [3 × 2]>
#> 10 <dbl [32]> carb <dbl [32]> <tibble [6 × 2]>
While this is decidedly not as pretty, it's capable of holding many types tidily. To extract the result above, just add %>% unnest(mpg_means). As-is, grouping variables are each held in a list element of group_values and in aggregated form in the first column of each mpg_means tibble.

Answer 2

When grouping your data within the map, you can rename the grouping variable to "level", since those values will form the column containing the levels of the grouping variable in the final data set.
When you have mixed types of grouping variables (e.g. both numeric and character), you'll also need to coerce the grouping variable to character in order to be able to bind the results together.
With those additions, you should get what you expect. (You can also skip the bind_rows by using map_df instead of map, to save a little bit of code, like I've done below.)
reprex::reprex_info()
#> Created by the reprex package v0.1.1.9000 on 2018-02-09
library(purrr)
library(dplyr)
data <- iris
vars <- names(data)
set_names(vars) %>%
map_df(function(var) {
var <- set_names(var, "level")
data %>%
group_by_at(var) %>%
summarize_at("Sepal.Length", "mean") %>%
mutate_at("level", as.character)
}, .id = "variable")
#> # A tibble: 126 x 3
#> variable level Sepal.Length
#> <chr> <chr> <dbl>
#> 1 Sepal.Length 4.3 4.3
#> 2 Sepal.Length 4.4 4.4
#> 3 Sepal.Length 4.5 4.5
#> 4 Sepal.Length 4.6 4.6
#> 5 Sepal.Length 4.7 4.7
#> 6 Sepal.Length 4.8 4.8
#> 7 Sepal.Length 4.9 4.9
#> 8 Sepal.Length 5 5.0
#> 9 Sepal.Length 5.1 5.1
#> 10 Sepal.Length 5.2 5.2
#> # ... with 116 more rows
You could also wrap the process in a function, and allow multiple variables to summarise with multiple functions. You'd have to spend a moment to come up with an evocative name though (I cheated and just used foo here).
foo <- function(data, vars, funs) {
grps <- names(data)
set_names(grps) %>%
map_df(function(grp) {
grp <- set_names(grp, "level")
data %>%
group_by_at(grp) %>%
summarize_at(vars, funs) %>%
mutate_at("level", as.character)
}, .id = "variable")
}
foo(iris, vars(Sepal.Length, Sepal.Width), funs(mean, sd))
#> # A tibble: 126 x 6
#> variable level Sepal.Length_mean Sepal.Width_mean Sepal.Length_sd
#> <chr> <chr> <dbl> <dbl> <dbl>
#> 1 Sepal.Length 4.3 4.3 3.000000 NaN
#> 2 Sepal.Length 4.4 4.4 3.033333 0
#> 3 Sepal.Length 4.5 4.5 2.300000 NaN
#> 4 Sepal.Length 4.6 4.6 3.325000 0
#> 5 Sepal.Length 4.7 4.7 3.200000 0
#> 6 Sepal.Length 4.8 4.8 3.180000 0
#> 7 Sepal.Length 4.9 4.9 2.950000 0
#> 8 Sepal.Length 5 5.0 3.120000 0
#> 9 Sepal.Length 5.1 5.1 3.477778 0
#> 10 Sepal.Length 5.2 5.2 3.425000 0
#> # ... with 116 more rows, and 1 more variables: Sepal.Width_sd <dbl>