How to delete rows for leading and trailing NAs by group in R

Question

I need to delete rows containing NA, but only if they are leading(trailing) i.e. before(after) any data appears for a variable. This is very similar to: How to find (not replace) leading NAs, gaps, and final NAs in data.table columns by category and: How delete leading and trailing rows by condition in R?

But, I need to do this process grouped by the variable "ID". I am going to impute the data of the NAs in between in a later step.

The same should apply to the trailing NAs.

The initial data.frame looks like this:

df1<-data.frame(ID=(rep(c("C1001","C1008","C1009","C1012"),each=17)),
  Year=(rep(c(1996:2012),4)),x1=(floor(runif(68,20,75))),x2= 
  (floor(runif(68,1,100))))

#Introduce leading / tailing NAs

df1[1:5,3]<-NA
df1[18:23,4]<-NA
df1[35:42,4]<-NA
df1[49:51,3]<-NA
df1[66:68,3]<-NA


#introduce "gap"- NAs
set.seed(123)
df1$x1[rbinom(68,1,0.1)==1]<-NA
df1$x2[rbinom(68,1,0.1)==1]<-NA

The output is quite long. This is to make a proper distinction between "gap"-NAs and "leading/trailing" NAs possible

head(df1,10)

      ID Year x1 x2
1  C1001 1996 NA 40
2  C1001 1997 NA 88
3  C1001 1998 NA 37
4  C1001 1999 NA 29
5  C1001 2000 NA 17
6  C1001 2001 42 18
7  C1001 2002 20 48
8  C1001 2003 30 26
9  C1001 2004 66 22
10 C1001 2005 32 67

The output should get rid of the leading NAs (see row 1:5 above) by group of ID. Or rows 18:23 in the example below:

df1[18:28,]

      ID Year x1 x2
18 C1008 1996 33 NA
19 C1008 1997 26 NA
20 C1008 1998 NA NA
21 C1008 1999 51 NA
22 C1008 2000 31 NA
23 C1008 2001 44 NA
24 C1008 2002 NA 56
25 C1008 2003 47 70
26 C1008 2004 39 91
27 C1008 2005 55 62
28 C1008 2006 40 43

The final output should look like this (depending on the random NAs thrown in of course!):

      ID Year x1 x2
6  C1001 2001 42 18
7  C1001 2002 20 48
8  C1001 2003 30 26
9  C1001 2004 66 22
10 C1001 2005 32 67
11 C1001 2006 NA  5
12 C1001 2007 24 70
13 C1001 2008 33 35
14 C1001 2009 60 41
15 C1001 2010 66 82
16 C1001 2011 47 91
17 C1001 2012 41 28
24 C1008 2002 NA 56
25 C1008 2003 47 70
26 C1008 2004 39 91
27 C1008 2005 55 62
28 C1008 2006 40 43
29 C1008 2007 39 54
30 C1008 2008 49  6
31 C1008 2009 NA 26
32 C1008 2010 NA 40
33 C1008 2011 42 20
34 C1008 2012 34 83
44 C1009 2005 51 96
45 C1009 2006 66 96
46 C1009 2007 37 NA
47 C1009 2008 58 26
48 C1009 2009 34 22
52 C1012 1996 51 78
53 C1012 1997 70 17
54 C1012 1998 69 41
55 C1012 1999 35 47
56 C1012 2000 37 86
57 C1012 2001 74 92
58 C1012 2002 54 NA
59 C1012 2003 71 67
60 C1012 2004 45 95
61 C1012 2005 42 52
62 C1012 2006 56 58
63 C1012 2007 28 34
64 C1012 2008 51 35
65 C1012 2009 33  2

Thanks a bunch!

Answer 1

Here is an approach that uses filter_at() that identifies leading NA values with cumsum() and trailing ones with the same idea but with the vector reversed.

library(dplyr)

df1 %>%
  group_by(ID) %>%
  filter_at(vars(-ID, -Year), all_vars(pmin(cumsum(!is.na(.)), rev(cumsum(!is.na(rev(.))))) != 0))

# A tibble: 42 x 4
# Groups:   ID [4]
   ID     Year    x1    x2
   <fct> <int> <dbl> <dbl>
 1 C1001  2001    42    18
 2 C1001  2002    20    48
 3 C1001  2003    30    26
 4 C1001  2004    66    22
 5 C1001  2005    32    67
 6 C1001  2006    NA     5
 7 C1001  2007    24    70
 8 C1001  2008    33    35
 9 C1001  2009    60    41
10 C1001  2010    66    82
# ... with 32 more rows

Answer 2

Here's a data.table solution that relies on rleid to only remove the leading NAs:

library(data.table)
dt <- as.data.table(df1)

dt[,
   .SD[!(rleid(x1) %in% c(1, max(rleid(x1))) & is.na(x1)) &
         !(rleid(x2) %in% c(1, max(rleid(x2))) & is.na(x2))],
   by = ID
   ]

To do it with multiple columns automatically, assuming that they all start with x you can do this:

dt[dt[, Reduce('&',
               lapply(.SD, function(x) !(rleid(x) %in% c(1, max(rleid(x1))) & is.na(x)))),
      by = ID,
      .SDcols = grep('x', names(dt))]$V1
   ]
# or using .SD as before

dt[,
   .SD[Reduce('&', lapply(.SD, function(x) !(rleid(x) %in% c(1, max(rleid(x1))) & is.na(x)))),
       .SDcols = grep('x', names(dt))],
   by = ID
   ]

Or same idea with dplyr:

library(dplyr)
library(data.table)

df1%>%
  group_by(ID)%>%
  filter_at(vars(starts_with('x')), all_vars(!(is.na(.) & rleid(.) %in% c(1, max(rleid(.))))))

Results in 42 rows:

# A tibble: 42 x 4
# Groups:   ID [4]
   ID     Year    x1    x2
   <fct> <int> <dbl> <dbl>
 1 C1001  2001    25    54
 2 C1001  2002    28    50
 3 C1001  2003    35    94
 4 C1001  2004    52    34
 5 C1001  2005    60    47
 6 C1001  2006    NA     9
 7 C1001  2007    67    86
 8 C1001  2008    58    40
 9 C1001  2009    61    73
10 C1001  2010    28    18
# ... with 32 more rows