R data.table 在组内查找最小值和最大值

Question

Hi I have this problem where I have a dataset where a person works in different companies. Now I want to find the duration of each company he worked for. Some person goes back to his previous company to work. Here is my dataset and my implementation, but it doesn't work when he goes back to his previous company later.

library(data.table)
data <- data.table(person = c(1,1,1,1,1,1,1,1), company = c(1,1,1,2,2,2,1,1),
               year = c(1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997))

You see person == 1 works in company 1 from 1990 to 1992 and then switched to company 2 from 1993 to 1995. Then he goes back to company 1 from 1996 to 1997.

I thought about using

min <- data[data[, .I[year == min(year)], by=.(person, company)]$V1]
setnames(min, "year", "start")

max <- data[data[, .I[year == max(year)], by=.(person, company)]$V1]
setnames(max, "year", "end")

duration <- merge(min, max, all = T)

which you get:

person company start  end
     1       1  1990 1997
     1       2  1993 1995

But what I want is:

person company start  end
     1       1  1990 1992
     1       2  1993 1995
     1       1  1996 1997

Any idea how to get that?

Thanks.

Answer 1

We can use rleid as a grouping variable

library(data.table)
data[, .(start =  min(year), end = max(year)),
    .(person, grp = rleid(company), company)][, grp := NULL][]

-output

   person company start  end
1:      1       1  1990 1992
2:      1       2  1993 1995
3:      1       1  1996 1997

---

Or may also use collapse

library(collapse)
data[, grp := rleid(company)]
collap(data,  ~ person + company + grp, list(fmin, fmax))
   person company fmin.year fmax.year grp
1:      1       1      1990      1992   1
2:      1       1      1996      1997   3
3:      1       2      1993      1995   2

Answer 2

There are probably better ways to do it, but here it goes:

library(data.table)
data = data.table(person = c(1,1,1,1,1,1,1,1), company = c(1,1,1,2,2,2,1,1),
                   year = c(1990, 1991, 1992, 1993, 1994, 1995, 1996, 1997))

data[, c('start', 'end', 'group') := 0]
group_count = 0

for (i in seq_len(nrow(data))) {
  if (i == 1) {
    next
    } else if (data[i, company] != data[i-1, company]) {
    group_count = group_count + 1
    data[i, group := group_count]
    } else {
    data[i, group := group_count]
    }
}

data[, c('start', 'end') := .(min(year), max(year)), by = group]

data = unique(data[, .(person, company, start, end)])

> data
   person company start  end
1:      1       1  1990 1992
2:      1       2  1993 1995
3:      1       1  1996 1997

Answer 3

Adopting @akrun's answer

If your dataset is large

data[, grp := rleid(company), by=.(person)]

min <- data[data[, .I[year == min(year)], by=.(person, company, grp)]$V1]
setnames(min, "year", "start")

max <- data[data[, .I[year == max(year)], by=.(person, company, grp)]$V1]
setnames(max, "year", "end")

duration <- merge(min, max, all = T)