在R中，使用dplyr中的“do”函数应用一系列复杂函数

Question

I would like to know that if is possible apply a sequence of especial functions using pipe chaining. For example, supose that I have the following data with 3 levels in group_var:

head(df)
 v1  v2 .... v9  group_var
  1   2       0          1
  5   3       2          0 
  2   1       3          1 
  1   8       9          2
  7   6       0          1
  5   9       2          0 

My firs question: I wish do the following

res<- df %>% group_by(group_var) %>% do( out = special_function(.) )  

where the special function has diferents components functions for each group. That is, in pseudocode, the special function is

special_function = (f0,f1,f2) 

so that f1, f2, f3 are mutually diferents. For example

f0<- function(data) apply(data, 2, min)
f1<- function(data) t(data)
f2<- function(data) as.list(data)

and

df[df$group_var == i ,] 

is the input of the fi function for each i = 0,1,2 . That is

res$out[[1]] ==  f0(df[df$group_var == 0 ,])
> T 
res$out[[2]] ==  f0(df[df$group_var == 1 ,])
> T
res$out[[3]] ==  f0(df[df$group_var == 2 ,])
> T

My second question is related to my first question. If the response of the first question is yes, I would like apply more operations. Actually, I want use the outputs of each fi and apply operations one related to other. For example

other_special_function<- function(d1,d2,d3)
{
ret<- cbind(d1 , d2)
ret2<- cbind(ret, apply(d,2,sum))
return(ret2)
}

res2<- res %>% do(out2 = other_special_function(.) )

Which

res2$out2 = other_special_function(res$out[[1]], res$out[[2]] , res$out[[3]] ) 

Answer 1

You absolutely can do this with pipes and functions; it is very flexible. If you have a function in mind that might be more complicated, as you describe, I would recommend going with purrr and map() rather than do() from dplyr.

Let's say you have everybody's favorite iris dataset. You can start out by using nest() from tidyr so that you have one row per group, with the data nested into mini-dataframes in a list-column.

library(dplyr)
library(tidyr)

iris %>% 
    nest(-Species) 

#> # A tibble: 3 × 2
#>      Species              data
#>       <fctr>            <list>
#> 1     setosa <tibble [50 × 4]>
#> 2 versicolor <tibble [50 × 4]>
#> 3  virginica <tibble [50 × 4]>

Then say you have some arbitrary function that you want to apply to each little mini-dataframe, and it is different depending on what the species is. This could be as complicated as you like.

fun <- function(df, species) {
    if (species == "setosa") {
        t(df)
    } else {
        df
    }
} 

You can use map2() from purrr to apply this function to each species/data pair, and you can put it in a new column in the data frame, and keep going, if you'd like.

library(purrr)

iris %>% 
    nest(-Species) %>%
    mutate(output = map2(data, Species, fun))

#> # A tibble: 3 × 3
#>      Species              data            output
#>       <fctr>            <list>            <list>
#> 1     setosa <tibble [50 × 4]>    <dbl [4 × 50]>
#> 2 versicolor <tibble [50 × 4]> <tibble [50 × 4]>
#> 3  virginica <tibble [50 × 4]> <tibble [50 × 4]>

This is a flexible technique for applying arbitrary operations to data, and is especially great for modeling .