Why does map %>% as.data.frame give different result than map_df? - r

Question

I'm new to R and trying to understand how to use map and map_df. Consider the following:
iris %>% split(.$Species) %>% map_df(function (x) apply(x[, 1:4], 2, mean))
And compare to
iris %>% split(.$Species) %>% map(function (x) apply(x[, 1:4], 2, mean)) %>% as.data.frame
The former gives the following output:
Sepal.Length Sepal.Width Petal.Length Petal.Width
<dbl> <dbl> <dbl> <dbl>
1 5.01 3.43 1.46 0.246
2 5.94 2.77 4.26 1.33
3 6.59 2.97 5.55 2.03
The latter gives:
setosa versicolor virginica
Sepal.Length 5.006 5.936 6.588
Sepal.Width 3.428 2.770 2.974
Petal.Length 1.462 4.260 5.552
Petal.Width 0.246 1.326 2.026
My question is: why? I would expect these two commands to give the same output. How can I get the second output with map_df function?

Answer 1

map_df() seems to bind the list elements by rows (same as map_dfr()) although it is not explicitly stated in its documentation. If you would like to bind by column, use map_dfc() instead. Note that the output is a tibble in which the use of rownames is discouraged. This document suggests ways to work with rownames in tibbles.
iris %>%
split(.$Species) %>%
map_dfc(function (x) apply(x[, 1:4], 2, mean))
# # A tibble: 4 x 3
# setosa versicolor virginica
# <dbl> <dbl> <dbl>
# 1 5.01 5.94 6.59
# 2 3.43 2.77 2.97
# 3 1.46 4.26 5.55
# 4 0.246 1.33 2.03

Answer 2

Firstly, you have used split instead of group_split which shows that you are creating a named list instead.
secondly, after you have created a named list (having three items), you are taking mean of first columns in all three items in list using apply
After the mean calculation, the result that remains is again a three item list with a single row containing mean of all four columns.
Now here comes the difference-
As you have created a named list in first step, if you use map_df or map_dfr (the result will b same in both), map function just row binds the output.
However, using map leaves the list as a list (named) which as.data.frame binds column wise.
library(tidyverse)
#1st case
iris %>% split(.$Species) %>% map_dfr(function (x) apply(x[, 1:4], 2, mean))
#> # A tibble: 3 x 4
#> Sepal.Length Sepal.Width Petal.Length Petal.Width
#> <dbl> <dbl> <dbl> <dbl>
#> 1 5.01 3.43 1.46 0.246
#> 2 5.94 2.77 4.26 1.33
#> 3 6.59 2.97 5.55 2.03
#OR
iris %>% split(.$Species) %>% map_df(function (x) apply(x[, 1:4], 2, mean))
#> # A tibble: 3 x 4
#> Sepal.Length Sepal.Width Petal.Length Petal.Width
#> <dbl> <dbl> <dbl> <dbl>
#> 1 5.01 3.43 1.46 0.246
#> 2 5.94 2.77 4.26 1.33
#> 3 6.59 2.97 5.55 2.03
#2nd case
iris %>% split(.$Species) %>% map(function (x) apply(x[, 1:4], 2, mean)) %>% as.data.frame
#> setosa versicolor virginica
#> Sepal.Length 5.006 5.936 6.588
#> Sepal.Width 3.428 2.770 2.974
#> Petal.Length 1.462 4.260 5.552
#> Petal.Width 0.246 1.326 2.026
#OR (with a slight difference of dropping row names)
iris %>% split(.$Species) %>% map_dfc(function (x) apply(x[, 1:4], 2, mean))
#> # A tibble: 4 x 3
#> setosa versicolor virginica
#> <dbl> <dbl> <dbl>
#> 1 5.01 5.94 6.59
#> 2 3.43 2.77 2.97
#> 3 1.46 4.26 5.55
#> 4 0.246 1.33 2.03
#If group_split would have been used instead
iris %>% group_split(Species) %>% map(function (x) apply(x[, 1:4], 2, mean)) %>% as.data.frame
#> c.Sepal.Length...5.006..Sepal.Width...3.428..Petal.Length...1.462..
#> Sepal.Length 5.006
#> Sepal.Width 3.428
#> Petal.Length 1.462
#> Petal.Width 0.246
#> c.Sepal.Length...5.936..Sepal.Width...2.77..Petal.Length...4.26..
#> Sepal.Length 5.936
#> Sepal.Width 2.770
#> Petal.Length 4.260
#> Petal.Width 1.326
#> c.Sepal.Length...6.588..Sepal.Width...2.974..Petal.Length...5.552..
#> Sepal.Length 6.588
#> Sepal.Width 2.974
#> Petal.Length 5.552
#> Petal.Width 2.026
#OR
iris %>% group_split(Species) %>% map_dfc(function (x) apply(x[, 1:4], 2, mean))
#> New names:
#> * NA -> ...1
#> * NA -> ...2
#> * NA -> ...3
#> # A tibble: 4 x 3
#> ...1 ...2 ...3
#> <dbl> <dbl> <dbl>
#> 1 5.01 5.94 6.59
#> 2 3.43 2.77 2.97
#> 3 1.46 4.26 5.55
#> 4 0.246 1.33 2.03
Thus, default binding of list into a dataframe is columnwise and if list is named, item names will be used as output df's column names. On the other hand if you are giving specific command through map_df or map_dfr then list items will be bind row-wise and name of list items are therefore not required.
iris %>% group_split(Species) %>% map_dfr(function (x) apply(x[, 1:4], 2, mean))
# A tibble: 3 x 4
Sepal.Length Sepal.Width Petal.Length Petal.Width
<dbl> <dbl> <dbl> <dbl>
1 5.01 3.43 1.46 0.246
2 5.94 2.77 4.26 1.33
3 6.59 2.97 5.55 2.03
Hope this is clear.