how to extract column names based on a condition? - r

Question

Consider this simple example
mytest <- data_frame(group = c('a', 'a', 'a', 'b', 'b', 'b'),
x = c(NA,NA,NA,5,6,7),
other_var = c(NA, NA, NA, 1,2,3),
y = c(3,5,6,NA,NA,NA),
another_var = c(1,2,3, NA,NA,NA),
label_x = c('hello','hello','hello','world','world','world'),
label_y =c('bada','bada','bada','boom','boom','boom'),
label_other_var = c('ak','ak','ak','run','run','run'),
label_another_var = c('noo','noo','noo','bie','bie','bie'))
# A tibble: 6 x 9
group x other_var y another_var label_x label_y label_other_var label_another_var
<chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr> <chr> <chr>
1 a NA NA 3 1 hello bada ak noo
2 a NA NA 5 2 hello bada ak noo
3 a NA NA 6 3 hello bada ak noo
4 b 5 1 NA NA world boom run bie
5 b 6 2 NA NA world boom run bie
6 b 7 3 NA NA world boom run bie
Here, I need to nest() this dataframe by group, and be able to extract the column names of the variables (in each nested dataframe) that are not NAs. The trick is that the actual name of the variable is shown in the label_ column
For instance, this is the output desired:
# A tibble: 4 x 2
group var
<chr> <chr>
1 a bada
2 a noo
3 b world
4 b run
Indeed, take group a. There is only one non-missing variables are y and another_var. However, the name of y is bada (as shown in the label_y variable) and the name of another_var is noo. Same reasoning for b.
I dont know how to do that with a map call after running
mytest %>% group_by(group) %>% nest()
# A tibble: 2 x 2
group data
<chr> <list>
1 a <tibble [3 x 8]>
2 b <tibble [3 x 8]>
Any ideas?
Thanks!
EDIT: the original, smaller, tibble proposed was the following
mytest <- data_frame(group = c('a', 'a', 'a', 'b', 'b', 'b'),
+ x = c(NA,NA,NA,5,6,7),
+ y = c(3,5,6,NA,NA,NA),
+ label_x = c('hello','hello','hello','world','world','world'),
+ label_y =c('bada','bada','bada','boom','boom','boom'))

Answer 1

After group by nesting, loop through the 'data' with map, summarise the the 'label' columns by extracting the first non-NA element, gather it to a single column while removing the NA (na.rm = TRUE)), select the 'var' column, and then do the unnest (after keeping only the columns of interest)
mytest %>%
group_by(group) %>%
nest %>%
mutate(var = map(data, ~
.x %>%
summarise(label_x = label_x[!is.na(x)][1],
label_y = label_y[!is.na(y)][1]) %>%
gather(key, var, na.rm = TRUE) %>%
select(var))) %>%
select(-data) %>%
unnest
# A tibble: 2 x 2#
# group var
# <chr> <chr>
#1 a bada
#2 b world
Update
If there are more columns, create unique column names and then loop through the corresponding column names with map2
nm1 <- unique(sub("label_", "", setdiff(names(mytest), "group")))
nm2 <- paste0("label_", nm1)
mytest %>%
group_by(group) %>%
nest %>%
mutate(var = map(data, ~
map2_chr(.x %>%
select(nm1),
.x %>%
select(nm2), ~
.y[!is.na(.x)][1]) %>%
na.omit %>%
tibble(var = .))) %>%
select(-data) %>%
unnest
# A tibble: 4 x 2
# group var
# <chr> <chr>
#1 a bada
#2 a noo
#3 b world
#4 b run

Answer 2

This will output the result you want:
mytest <- data_frame(group = c('a', 'a', 'a', 'b', 'b', 'b'),
x = c(NA,NA,NA,5,6,7),
y = c(3,5,6,NA,NA,NA),
label_x = c('hello','hello','hello','world','world','world'),
label_y =c('bada','bada','bada','boom','boom','boom'))
extract_good_colnames <- function(df, subgroup){
subset <- filter(df, group == subgroup)
if(sum(is.na(subset$x)) > 0){
colname = 'label_y'
}else if(sum(is.na(subset$y)) > 0){
colname = 'label_x'
}
return(tibble(group = subgroup, var = as.character(subset[1, colname])))
}
groups <- unique(mytest$group)
map_df(groups, function(x) extract_good_colnames(mytest, x))