Use a function name that's a string in map loop? - r

Question

Some code:
mymtcars <- mtcars %>% head %>% rownames_to_column('model') %>% group_by(vs) %>% nest
mymtcars
vs data
<dbl> <list>
1 0 <tibble [3 × 11]>
2 1 <tibble [3 × 11]>
I can fit a linear model on this list column df like so:
mymtcars %>%
+ mutate(mod = map(.x = data, ~ lm(.x$mpg ~ .x$cyl)))
# A tibble: 2 x 3
# Groups: vs [2]
vs data mod
<dbl> <list> <list>
1 0 <tibble [3 × 11]> <lm>
2 1 <tibble [3 × 11]> <lm>
What if my function name is a field?
mymtcars2 <- mtcars %>% head %>% rownames_to_column('model') %>% group_by(vs) %>% nest %>% crossing(func = c('lm'))
> mymtcars2
# A tibble: 2 x 3
vs data func
<dbl> <list> <chr>
1 0 <tibble [3 × 11]> lm
2 1 <tibble [3 × 11]> lm
I gave it a try with:
mymtcars2 %>%
+ mutate(mod = map2(.x = data, .y = func, ~ .y(.x$mpg ~ .x$cyl)))
Error: Problem with `mutate()` input `mod`.
x could not find function ".y"
ℹ Input `mod` is `map2(.x = data, .y = func, ~.y(.x$mpg ~ .x$cyl))`.
How can I pass the function to call in map and then call it in the above block?

Answer 1

May be using match.fun inside map2 like below:
models <- mymtcars2 %>%
mutate(mod = map2(.x = data, .y = func, ~ match.fun(.y)(.x$mpg ~ .x$cyl)))
Output:
[[1]]
Call:
match.fun(.y)(formula = .x$mpg ~ .x$cyl)
Coefficients:
(Intercept) .x$cyl
36.926733 -2.728218
[[2]]
Call:
match.fun(.y)(formula = .x$mpg ~ .x$cyl)
Coefficients:
(Intercept) .x$cyl
41.9400 -3.8025

Answer 2

I also found that I can use get:
mymtcars2 %>%
mutate(mod = map2(.x = data, .y = func, ~ get(.y)(.x$mpg ~ .x$cyl)))
Am unsure of when to use one over the other.

Answer 3

A different option could be:
mymtcars2 %>%
mutate(mod = map2(.x = data,
.y = func,
~ exec(.y, mpg ~ cyl, data = .x)))
vs data func mod
<dbl> <list> <chr> <list>
1 0 <tibble [3 × 11]> lm <lm>
2 1 <tibble [3 × 11]> lm <lm>

Answer 4

Since {dplyr} >= 1.0 this kind of problems can be solved with dplyr::rowwise. We can use it either with a classic do.call, in which case we have to wrap the arguments in list(), or with rlang::exec. With dlpyr::rowwise we don't need map2 which makes things more readable since there is no lambda function with .x .y. However, since the output column stores lm objects (and not an atomic vector), the result has to be wrapped in mod = list(...).
library(tidyverse)
mymtcars2 %>%
rowwise %>%
mutate(mod = list(do.call(func, list(mpg ~ cyl, data = data))))
#> # A tibble: 2 x 4
#> # Rowwise:
#> vs data func mod
#> <dbl> <list> <chr> <list>
#> 1 0 <tibble [3 × 11]> lm <lm>
#> 2 1 <tibble [3 × 11]> lm <lm>
mymtcars2 %>%
rowwise %>%
mutate(mod = list(exec(func, mpg ~ cyl, data = data)))
#> # A tibble: 2 x 4
#> # Rowwise:
#> vs data func mod
#> <dbl> <list> <chr> <list>
#> 1 0 <tibble [3 × 11]> lm <lm>
#> 2 1 <tibble [3 × 11]> lm <lm>
Created on 2021-08-28 by the reprex package (v0.3.0)