Ttykuu

I have a little nut to crack.

I have a data.frame like this:

 group criterium
1 A NA
2 A TRUE
3 A TRUE
4 A TRUE
5 A FALSE
6 A FALSE
7 A TRUE
8 A TRUE
9 A FALSE
10 A TRUE
11 A TRUE
12 A TRUE
13 B NA
14 B FALSE
15 B TRUE
16 B TRUE
17 B TRUE
18 B FALSE

structure(list(group = structure(c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L), .Label = c("A", 
"B"), class = "factor"), criterium = c(NA, TRUE, TRUE, TRUE, 
FALSE, FALSE, TRUE, TRUE, FALSE, TRUE, TRUE, TRUE, NA, FALSE, 
TRUE, TRUE, TRUE, FALSE)), class = "data.frame", row.names = c(NA, 
-18L))

And I want to rank the groups of TRUE in column criterium in ascending order while disregarding the FALSEand NA. The goal is to have a unique group identifier inside each group of group.

So the result should look like:

 group criterium goal
1 A NA NA
2 A TRUE 1
3 A TRUE 1
4 A TRUE 1
5 A FALSE NA
6 A FALSE NA
7 A TRUE 2
8 A TRUE 2
9 A FALSE NA
10 A TRUE 3
11 A TRUE 3
12 A TRUE 3
13 B NA NA
14 B FALSE NA
15 B TRUE 1
16 B TRUE 1
17 B TRUE 1
18 B FALSE NA

I'm sure there is a relatively easy way to do this, I just can't think of one. I experimented with dense_rank() and other window functions of dplyr, but to no avail.

Thanks for the help!

Maybe I have over-complicated this but one way with dplyr is

library(dplyr)

df %>%
 mutate(temp = replace(criterium, is.na(criterium), FALSE), 
 temp1 = cumsum(!temp)) %>%
 group_by(temp1) %>%
 mutate(goal = +(row_number() == which.max(temp) & any(temp))) %>%
 group_by(group) %>%
 mutate(goal = ifelse(temp, cumsum(goal), NA)) %>%
 select(-temp, -temp1)

# group criterium goal
# <fct> <lgl> <int>
# 1 A NA NA
# 2 A TRUE 1
# 3 A TRUE 1
# 4 A TRUE 1
# 5 A FALSE NA
# 6 A FALSE NA
# 7 A TRUE 2
# 8 A TRUE 2
# 9 A FALSE NA
#10 A TRUE 3
#11 A TRUE 3
#12 A TRUE 3
#13 B NA NA
#14 B FALSE NA
#15 B TRUE 1
#16 B TRUE 1
#17 B TRUE 1
#18 B FALSE NA

We first replace NAs in criterium column to FALSE and take cumulative sum over the negation of it (temp1). We group_by temp1 and assign 1 to every first TRUE value in the group. Finally grouping by group we take a cumulative sum for TRUE values or return NA for FALSE and NA values.

We can create a custom function via rle, and use it per group, i.e.

f1 <- function(x) 
 x[is.na(x)] <- FALSE
 rle1 <- rle(x)
 y <- rle1$values
 rle1$values[!y] <- 0
 rle1$values[y] <- cumsum(rle1$values[y])
 return(inverse.rle(rle1))



do.call(rbind, 
 lapply(split(df, df$group), function(i)!i$criterium, NA); 
 i))

Of course, If you want you can apply it via dplyr, i.e.

library(dplyr)

df %>% 
 group_by(group) %>% 
 mutate(goal = f1(criterium), 
 goal = replace(goal, is.na(criterium)|!criterium, NA))

which gives,

# A tibble: 18 x 3
# Groups: group [2]
 group criterium goal
 <fct> <lgl> <dbl>
 1 A NA NA
 2 A TRUE 1
 3 A TRUE 1
 4 A TRUE 1
 5 A FALSE NA
 6 A FALSE NA
 7 A TRUE 2
 8 A TRUE 2
 9 A FALSE NA
10 A TRUE 3
11 A TRUE 3
12 A TRUE 3
13 B NA NA
14 B FALSE NA
15 B TRUE 1
16 B TRUE 1
17 B TRUE 1
18 B FALSE NA

A different dplyr possibility using rle() and dense_rank() could be:

df %>%
 group_by(group) %>%
 mutate(rleid = with(rle(criterium), rep(seq_along(lengths), lengths))) %>%
 group_by(group, criterium) %>%
 mutate(res = dense_rank(rleid)) %>%
 ungroup() %>%
 mutate(res = ifelse(criterium, res, NA)) %>%
 select(-rleid)

 group criterium res
 <fct> <lgl> <int>
 1 A NA NA
 2 A TRUE 1
 3 A TRUE 1
 4 A TRUE 1
 5 A FALSE NA
 6 A FALSE NA
 7 A TRUE 2
 8 A TRUE 2
 9 A FALSE NA
10 A TRUE 3
11 A TRUE 3
12 A TRUE 3
13 B NA NA
14 B FALSE NA
15 B TRUE 1
16 B TRUE 1
17 B TRUE 1
18 B FALSE NA

A data.table option using rle

library(data.table)
DT <- as.data.table(dat)
DT[, goal := 
 r <- rle(replace(criterium, is.na(criterium), FALSE))
 r$values <- with(r, cumsum(values) * values) 
 out <- inverse.rle(r) 
 replace(out, out == 0, NA)
, by = group]
DT
# group criterium goal
# 1: A NA NA
# 2: A TRUE 1
# 3: A TRUE 1
# 4: A TRUE 1
# 5: A FALSE NA
# 6: A FALSE NA
# 7: A TRUE 2
# 8: A TRUE 2
# 9: A FALSE NA
#10: A TRUE 3
#11: A TRUE 3
#12: A TRUE 3
#13: B NA NA
#14: B FALSE NA
#15: B TRUE 1
#16: B TRUE 1
#17: B TRUE 1
#18: B FALSE NA

data

dat <- structure(list(group = structure(c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 2L), .Label = c("A", 
"B"), class = "factor"), criterium = c(NA, TRUE, TRUE, TRUE, 
FALSE, FALSE, TRUE, TRUE, FALSE, TRUE, TRUE, TRUE, NA, FALSE, 
TRUE, TRUE, TRUE, FALSE)), class = "data.frame", row.names = c(NA, 
-18L))