Check the previous row and find when the number change to zero in R

I have a data like this:

d <- data.frame(
  ab = c(3, 4, 2, 6),
  dur_1 = c(32, 1, 3, 4),
  dur_2 = c(27, 9, 26, 5),
  dur_3 = c(25, 8, 21, 48),
  dur_5 = c(0, 4, 0, 42),
  dur_6 = c(0, 0, 0, 0),
  dur_7 = c(0, 0, 0, 0),
  cd = c(45, 67, 34, 78)
)

What I want to do is make a new column pre_dur.
pre_dur is the last nonnegative value of previous rows for the variables starts with dur_.
So, to get pre_dur, scan through the dur_ variables in the previous row, find where the number starts to change zero. The very last nonnegative number is the value that I want.

My expected output should be like this:

d1<-data.frame(ab=c(3,4,2,6),
               dur_1=c(32,1,3,4),
               dur_2=c(27,9,26,5),
               dur_3=c(25,8,21,48),
               dur_5=c(0,4,0,42),
               dur_6=c(0,0,0,0),
               dur_7=c(0,0,0,0),
               cd=c(45,67,34,78),
               pre_dur=c(NA,25,4,21))

Actually the code below works:

d1 <- d %>%
  mutate(pre_dur = case_when(lag(dur_7) > 0 ~ lag(dur_7),
                             lag(dur_6) > 0 ~ lag(dur_6),
                             lag(dur_5) > 0 ~ lag(dur_5),
                             lag(dur_3) > 0 ~ lag(dur_3),
                             lag(dur_2) > 0 ~ lag(dur_2),
                             lag(dur_1) > 0 ~ lag(dur_1),
                             TRUE ~ NA_real_))

However, in my actual data, the natural number after dur_ can be changed. So I need a generalized code. How to do that?

How about this solution?

library(tidyverse)
temp <- d %>% 
  pivot_longer(-ab) %>% 
  group_by(ab) %>% 
  mutate(out = ifelse(value == 0, lag(value), 0)) %>% 
  filter(out != 0) 
  
# bind_cols(d,  pre_dur = c(NA, temp$out[2:nrow(temp)]))

bind_cols(d,  pre_dur = lag(temp$out))

This assumes that once a value drops to 0, then it doesn't increase. If this isn't the case, then the ifelse statement needs to capture this.

First, find the value for each group, then, lag it once. We could do it using rowwise(), but it's probably faster and more readable to transform it into long format and back again. I.e.

library(dplyr)
library(tidyr)

d |>
    pivot_longer(starts_with("dur")) |>
    mutate(pre_dur = last(value[value > 0]), .by = "ab") |>
    pivot_wider() |>
    mutate(pre_dur = lag(pre_dur))

# A tibble: 4 × 9
     ab    cd pre_dur dur_1 dur_2 dur_3 dur_5 dur_6 dur_7
  <dbl> <dbl>   <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1     3    45      NA    32    27    25     0     0     0
2     4    67      25     1     9     8     4     0     0
3     2    34       4     3    26    21     0     0     0
4     6    78      21     4     5    48    42     0     0

A way in base using max.col with last.

i <- which(startsWith(names(d), "dur"))
j <- max.col(d[i] > 0, "last")
cbind(d, pre_dur = c(NA, d[cbind(seq_len(nrow(d)), i[j])][-nrow(d)]))
#  ab dur_1 dur_2 dur_3 dur_5 dur_6 dur_7 cd pre_dur
#1  3    32    27    25     0     0     0 45      NA
#2  4     1     9     8     4     0     0 67      25
#3  2     3    26    21     0     0     0 34       4
#4  6     4     5    48    42     0     0 78      21