Can I combine a dplyr mutate_at & mutate_if statement?

Question

I have the following example output:

   country country-year year     a     b
1  France  France2000   2000       NA    NA 
2  France  France2001   2001     1000  1000  
3  France  France2002   2002       NA    NA
4  France  France2003   2003     1600  2200
5  France  France2004   2004       NA    NA
6  UK          UK2000   2000     1000  1000  
7  UK          UK2001   2001       NA    NA
8  UK          UK2002   2002     1000  1000  
9  UK          UK2003   2003       NA    NA
10 UK          UK2004   2004       NA    NA
11 Germany     UK2000   2000       NA    NA 
12 Germany     UK2001   2001       NA    NA
13 Germany     UK2002   2002       NA    NA  
14 Germany     UK2003   2003       NA    NA
15 Germany     UK2004   2004       NA    NA

I want to interpolate the data I (but not extrapolate), and remove the columns for which columns a and b are both NA. In other words I would like to remove all the columns for which I cannot interpolate; in the example:

1  France  France2000        NA    NA
5  France  France2004        NA    NA
9  UK          UK2003        NA    NA
10 UK          UK2004        NA    NA
11 Germany     UK2000        NA    NA 
12 Germany     UK2001        NA    NA
13 Germany     UK2002        NA    NA  
14 Germany     UK2003        NA    NA
15 Germany     UK2004        NA    NA

There are two options that almost do what I want:

library(tidyverse)
library(zoo)
df %>%
  group_by(country) %>%
  mutate_at(vars(a:b),~na.fill(.x,c(NA, "extend", NA))) %>% 
  filter(!is.na(a) | !is.na(b))

AND

df%>% 
  group_by(Country)%>% 
  mutate_if(is.numeric,~if(all(is.na(.x))) NA else na.fill(.x,"extend"))

Would it be possible to combine these codes, doing something like this:

df <- df%>%
  group_by(country)%>%
  mutate_at(vars(a:b),~if(all(is.na(.x))) NA else(.x,c(NA, "extend", NA)))
  filter(!is.na(df$a | df$a))

Desired output:

   country country-year    a     b 
2  France  France2001      1000  1000  
3  France  France2002      1300  1600
4  France  France2003      1600  2200
6  UK          UK2000      1000  1000  
7  UK          UK2001         0     0
8  UK          UK2002      1000  1000

Answer 1

I know this doesn't directly answer the question how to combine mutate_if and mutate_at, but this solves your general problem:

I first get rid of the countries where all a and b are missing, and then determine for each country the minimum and maximum Year, which is not missing. After filtering these, I use the na.fill.

library(dplyr)
library(readr)
library(zoo)

country_data %>% 
  mutate(Year = parse_number(`country-year`)) %>% 
  group_by(country) %>% 
  mutate(not_all_na = any(!(is.na(a) & is.na(b)))) %>% 
  filter(not_all_na) %>% 
  mutate(Year_min_not_na = min(Year[!(is.na(a) & is.na(b))]), 
         Year_max_not_na = max(Year[!(is.na(a) & is.na(b))])) %>% 
  filter(Year >= Year_min_not_na, Year <= Year_max_not_na) %>% 
  mutate_at(vars(a:b), ~na.fill(.x, "extend")) 

# A tibble: 6 x 8
# Groups:   country [2]
#   country `country-year`     a     b  Year not_all_na Year_min_not_na Year_max_not_na
#   <fct>   <fct>          <dbl> <dbl> <dbl> <lgl>                <dbl>           <dbl>
# 1 France  France2001      1000  1000  2001 TRUE                  2001            2003
# 2 France  France2002      1300  1600  2002 TRUE                  2001            2003
# 3 France  France2003      1600  2200  2003 TRUE                  2001            2003
# 4 UK      UK2000          1000  1000  2000 TRUE                  2000            2002
# 5 UK      UK2001          1000  1000  2001 TRUE                  2000            2002
# 6 UK      UK2002          1000  1000  2002 TRUE                  2000            2002

Data

country_data <- 
  structure(list(country = structure(c(1L, 1L, 1L, 1L, 1L, 3L, 3L, 3L, 3L, 3L, 2L, 2L, 2L, 2L, 2L), 
                                                   .Label = c("France", "Germany", "UK"), class = "factor"), 
                               country.year = structure(c(1L, 2L, 3L, 4L, 5L, 6L, 7L, 8L, 9L, 10L, 6L, 7L, 8L, 9L, 10L), 
                                                        .Label = c("France2000", "France2001", "France2002", "France2003", 
                                                                   "France2004", "UK2000", "UK2001", "UK2002", "UK2003", "UK2004"), 
                                                        class = "factor"), 
                               a = c(NA, 1000L, NA, 1600L, NA, 1000L, NA, 1000L, NA, NA, NA, NA, NA, NA, NA),
                               b = c(NA, 1000L, NA, 2200L, NA, 1000L, NA, 1000L, NA, NA, NA, NA, NA, NA, NA)), 
                          class = "data.frame", row.names = c(NA, -15L))

Answer 2

Unfortunately the solution of @kath onyl works in given example but fails if only one column contains data, eg:

country country-year year     a     b
France  France2000   2000       NA    NA 
France  France2001   2001     1000  1000  
France  France2002   2002       NA    NA
France  France2003   2003     1600  2200
France  France2004   2004       NA    NA
UK          UK2000   2000     1000  1000  
UK          UK2001   2001       NA    NA
UK          UK2002   2002     1000  1000  
UK          UK2003   2003       NA    NA
UK          UK2004   2004       NA    NA
Germany     UK2000   2000       NA    NA 
Germany     UK2001   2001       NA   500
Germany     UK2002   2002       NA    NA  
Germany     UK2003   2003       NA  1100
Germany     UK2004   2004       NA    NA

Unfortunately too, the answer to the OPs question is no, you can't mix mutate_at and mutate_if (there's no function that allows you to specify .predicate and .vars)

but you can use a predict function within the function used in mutate_at. So here is my solution using mutate_at containing a predict function:

df %>%
  group_by(country) %>%
  # Interpolate if at least two non-null values are present
  mutate_at(vars(a,b), funs(if(sum(!is.na(.))<2) {NA_real_} else{approx(year, ., year)$y})) %>% 
  # keep only rows with original or interpolated values in either column a or b
  filter_at(vars(a,b), any_vars(!is.na(.)))

Answer 3

With dplyr 0.8.3 inspired by:

• community.rstudio.com: How to mutate_at/mutate_if multiple columns using condition on other column outside .vars ?

library(dplyr)
(iris [1:3,] 
    %>% mutate_at(c("Petal.Width"),
                  list(~ifelse(Sepal.Width == 3.5, 
                               .+10,
                               .+100)
                  )
    )
)
#>   Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> 1          5.1         3.5          1.4        10.2  setosa
#> 2          4.9         3.0          1.4       100.2  setosa
#> 3          4.7         3.2          1.3       100.2  setosa

with a new column toto:

library(dplyr)
(iris [1:3,] 
  %>% mutate_at(c("Petal.Width"),
                list(toto=~ifelse(Sepal.Width == 3.5, 
                             .+10,
                             .+100)
                )
  )
)
#>   Sepal.Length Sepal.Width Petal.Length Petal.Width Species  toto
#> 1          5.1         3.5          1.4         0.2  setosa  10.2
#> 2          4.9         3.0          1.4         0.2  setosa 100.2
#> 3          4.7         3.2          1.3         0.2  setosa 100.2

^{Created on 2019-07-30 by the reprex package (v0.2.1)}