R-遍历每个组并动态分配值

Question

I have the following dataset:

id	row_name	start_date	end_date	rows_overlap_period
person_1	1	2010-04-23	2010-06-22	2,3,4,5,6
person_1	2	2010-04-25	2010-06-24	3,4,5,6
person_1	3	2010-04-27	2010-06-26	4,5,6,7
person_1	4	2010-04-29	2010-06-28	5,6,7,8
person_1	5	2010-04-30	2010-06-29	6,7,8
person_1	6	2010-05-08	2010-07-07	7,8
person_1	7	2010-06-26	2010-08-25	8
person_1	8	2010-06-28	2010-08-27
person_2	9	2010-07-30	2010-09-28	10
person_2	10	2010-08-02	2010-10-01

The "rows_overlap_period" column indicates which other records started between the 'start_date' and 'end_date' period.

However, I would iterate within each group to arrive at the following result:

id	row_name	start_date	end_date	rows_overlap_period
person_1	1	2010-04-23	2010-06-22	2,3,4,5,6
person_1	2	2010-04-25	2010-06-24
person_1	3	2010-04-27	2010-06-26
person_1	4	2010-04-29	2010-06-28
person_1	5	2010-04-30	2010-06-29
person_1	6	2010-05-08	2010-07-07
person_1	7	2010-06-26	2010-08-25	8
person_1	8	2010-06-28	2010-08-27
person_2	9	2010-07-30	2010-09-28	10
person_2	10	2010-08-02	2010-10-01

This "output" would be the result of the 'following algorithm':

For each group:

1. Get the first row for which the 'rows_overlap_period' column are not empty (eg: row_name = 1)
2. For selected row, I get list of overlap values (eg "2,3,4,5,6") and assign ' ' to all row_names in column 'rows_overlap_period' (in this case, replace the values "3,4,5,6", "4,5,6,7", "5,6,7,8", "6,7,8" and "7,8" with " ")
3. For the same group, I look for the next line that doesn't have null values and repeat steps 1 and 2. If it doesn't exist, I move on to the next group.

Reproducible example (what I got so far):

# Input data
data.frame(id = c("person_1", "person_1", "person_1", "person_1", "person_1",
                     "person_1", "person_1", "person_1", "person_2",
                     "person_2"),
           row_name = rep(1:10),
           start_date = as.Date(c("2010-04-23", "2010-04-25", "2010-04-27",
                                  "2010-04-29", "2010-04-30", "2010-05-08",
                                  "2010-06-26", "2010-06-28", "2010-07-30",
                                  "2010-08-02")),
           end_date = as.Date(c("2010-06-22", "2010-06-24", "2010-06-26",
                                "2010-06-28", "2010-06-29", "2010-07-07",
                                "2010-08-25", "2010-08-27", "2010-09-28",
                                "2010-10-01"))) -> data


# Find overlaps (column rows_overlap_period)
sqldf::sqldf("select a.*,
                     coalesce(group_concat(b.row_name), ' ') as rows_overlap_period
             from data a
             left join data b on
                       a.id = b.id and
                       not a.row_name = b.row_name and
                       (b.start_date between
                        a.start_date and a.end_date) 
                    group by a.rowid
                    order by a.rowid") -> data

I was really trying to find some solution using dplyr, data.table or sqldf directly, but I can't find ways not to implement 'loops within loops' - which would degrade performance a lot.

Does anyone have any suggestions on how I can reach this?

Answer 1

We could create a grouping column to do this in addition to the 'id' column

library(dplyr)
data %>% 
  group_by(id) %>% 
  mutate(grp = cumsum(lead(!nzchar(trimws(rows_overlap_period)),
     default = FALSE))) %>% 
  group_by(grp, .add = TRUE) %>% 
  mutate(rows_overlap_period = case_when(row_number() ==1 ~ 
      rows_overlap_period, TRUE ~ "")) %>%
  ungroup %>% 
  select(-grp)

-output

# A tibble: 10 × 5
   id       row_name start_date end_date   rows_overlap_period
   <chr>       <int> <date>     <date>     <chr>              
 1 person_1        1 2010-04-23 2010-06-22 "2,3,4,5,6"        
 2 person_1        2 2010-04-25 2010-06-24 ""                 
 3 person_1        3 2010-04-27 2010-06-26 ""                 
 4 person_1        4 2010-04-29 2010-06-28 ""                 
 5 person_1        5 2010-04-30 2010-06-29 ""                 
 6 person_1        6 2010-05-08 2010-07-07 ""                 
 7 person_1        7 2010-06-26 2010-08-25 "8"                
 8 person_1        8 2010-06-28 2010-08-27 ""                 
 9 person_2        9 2010-07-30 2010-09-28 "10"               
10 person_2       10 2010-08-02 2010-10-01 ""