Gfrktyl

I have a data frame like this:

df <- data.frame(id = c("A", "A", "A", "A", "A", "A", "A", "A", 

                    "B", "B", "B", "B", "B", "B"),

             var1 = c("100", "200", "300", NA, NA, NA, NA, NA,

                      "100", "200", "300", NA, NA, NA), 

             var2 = c("100", NA, NA, "400", "500", "600", NA, NA,

                      NA, NA, NA, "400", NA, NA),

             var3 = c("200", NA, NA, NA, NA, NA, "700", "800",

                      "500", NA, NA, NA, "500", "600"))

which looks like this:

  id var1 var2 var3

   A  100  100  200

   A  200 <NA> <NA>

   A  300 <NA> <NA>

   A <NA>  400 <NA>

   A <NA>  500 <NA>

   A <NA>  600 <NA>

   A <NA> <NA>  700

   A <NA> <NA>  800

   B  100 <NA>  500

   B  200 <NA> <NA>

   B  300 <NA> <NA>

   B <NA>  400 <NA>

   B <NA> <NA>  500

   B <NA> <NA>  600

I would like to shift values in columns up if there are missing values above (by group). The result should look like this:

  id var1 var2 var3

   A  100  100  200

   A  200  400  700

   A  300  500  800

   A <NA>  600 <NA>

   B  100  400  500

   B  200 <NA>  500

   B  300 <NA>  600

I have no idea how to do this. Any thoughts?

Here is a rough concept using data.table that can be refined:

library(data.table)

# Helper function:

shift_up <- function(x) {

  n <- length(x)

  x <- x[!is.na(x)]

  length(x) <- n

  x

}



setDT(df)

df[, lapply(.SD, shift_up), id][!(is.na(var1) & is.na(var2) & is.na(var3))]



   id var1 var2 var3

1:  A  100  100  200

2:  A  200  400  700

3:  A  300  500  800

4:  A <NA>  600 <NA>

5:  B  100  400  500

6:  B  200 <NA>  500

7:  B  300 <NA>  600

Don't think this is the most efficient way to do it but one option

library(rowr)



df1 <- do.call(rbind, lapply(split(df, df$id), function(x) {

    data.frame(id = x$id[1], do.call(cbind.fill,c(sapply(x[-1], na.omit),fill = NA)))

}))

names(df1) <- names(df)

df1





#    id   var1   var2   var3

#A.1  A    100    100    200

#A.2  A    200    400    700

#A.3  A    300    500    800

#A.4  A   <NA>    600   <NA>

#B.1  B    100    400    500

#B.2  B    200   <NA>    500

#B.3  B    300   <NA>    600

We split the dataframe into list of dataframe for every id and for each dataframe we remove the NA values using na.omit and use cbind.fill to fill the values with NA and finally merge the list of dataframes back into one using rbind with do.call.

Here is an option with data.table. Convert the 'data.frame' to 'data.table' (setDT(df)), grouped by 'id', order the other column based on the NA values, then create an index to remove the rows where all the elements are NA

library(data.table)

df1 <- setDT(df)[,  lapply(.SD, function(x) x[order(is.na(x))]), id]

df1[df1[,!Reduce(`&`, lapply(.SD, is.na)), .SDcols = var1:var3]]

#   id var1 var2 var3

#1:  A  100  100  200

#2:  A  200  400  700

#3:  A  300  500  800

#4:  A <NA>  600 <NA>

#5:  B  100  400  500

#6:  B  200 <NA>  500

#7:  B  300 <NA>  600

Or using the same logic with tidyverse. Grouped by 'id', change the order or elements in all other column with mutate_all by ordering on the logical vector (is.na(column)) and keep the rows having at least one non-NA (filter_at)

library(tidyverse)

df %>% 

   group_by(id) %>% 

   mutate_all(funs(.[order(is.na(.))])) %>% 

   filter_at(vars(var1:var3), any_vars(!is.na(.)))

# A tibble: 7 x 4

# Groups:   id [2]

#  id    var1  var2  var3 

#  <fct> <fct> <fct> <fct>

#1 A     100   100   200  

#2 A     200   400   700  

#3 A     300   500   800  

#4 A     <NA>  600   <NA> 

#5 B     100   400   500  

#6 B     200   <NA>  500  

#7 B     300   <NA>  600  

Ordering a vector/column based on logical indexing is simple.

v1 <- c(1:3, NA, 5, NA, 7)

order(is.na(v1)) #gives the index of order

#[1] 1 2 3 5 7 4 6

use that index to change the order of values

v1[order(is.na(v1))]

#[1]  1  2  3  5  7 NA NA