I have a data.frame / data.table in R as follows:
df <- data.frame(
ID = c(rep("A", 20)),
year = c(1968, 1971, 1972, 1973, 1974, 1976, 1978, 1980, 1982, 1984, 1985,
1986, 1987, 1988, 1990, 1991, 1992, 1993, 1994, 1995)
)
I'd like to subset the df in order to keep only those entries which have at least five consecutive years. In this example this is the case in two periods (1984:1988 and 1990:1995). How can I do this in R?
A compact solution using diff
and cumsum
:
setDT(df)[, grp := cumsum(c(0, diff(year)) > 1), by = ID
][, if (.N > 4) .SD, by = .(ID, grp)][, grp := NULL][]
which gives the desired result:
ID year
1: A 1984
2: A 1985
3: A 1986
4: A 1987
5: A 1988
6: A 1990
7: A 1991
8: A 1992
9: A 1993
10: A 1994
11: A 1995
Explanation:
grp := cumsum(c(0, diff(year)) > 1), by = ID
you create a (temporary) grouping variable for consecutive years for each ID
.if (.N > 4) .SD, by = .(ID, grp)
you select only groups with 5 or more consecutive years for each ID
.grp := NULL
you remove the (temporary) grouping variable.A compareble approach in base R:
i <- with(df, ave(year, ID, FUN = function(x) {
r <- rle(cumsum(c(0, diff(year)) > 1));
rep(r$lengths, r$lengths)
} ))
df[i > 4,] # or df[which(i > 4),]
which will get you the same result.
Here is another way:
df2 <- NULL
sapply(seq(nrow(df)), function(x)
{
ifelse((sum(diff(df[x:(x+4), "year"], 1)) == 4 &
sum(diff(df[x:(x+4), "year"], 1) == 1) == 4),
df2 <<- rbind(df2, df[x:(x+4),]),"")
})
df2 <- unique(df2)
We can try
i1 <- with(df, as.logical(ave(year, ID, FUN = function(x) {
i1 <- (x[-1] - x[-length(x)]) ==1
i2 <- c(FALSE, i1)
i3 <- c(i1, FALSE)
rl <- rle(i2|i3)
rl$values[rl$values][rl$lengths[rl$values] <5] <- FALSE
rep(rl$values, rl$lengths)
})))
df[i1,]
# ID year
#10 A 1984
#11 A 1985
#12 A 1986
#13 A 1987
#14 A 1988
#15 A 1990
#16 A 1991
#17 A 1992
#18 A 1993
#19 A 1994
#20 A 1995
Or use data.table
library(data.table)
i1 <- setDT(df)[, ind := (year - shift(year, fill= year[1L]))==1L , ID][,
{i1 <- .I[.N * ind > 3]
.(v1 = head(i1,1)-1, v2 = tail(i1, 1))},
.(ID, rl = rleid(ind))][, seq(v1, v2) , rl]$V1
df[, ind := NULL][i1]
# ID year
# 1: A 1984
# 2: A 1985
# 3: A 1986
# 4: A 1987
# 5: A 1988
# 6: A 1990
# 7: A 1991
# 8: A 1992
# 9: A 1993
#10: A 1994
#11: A 1995
Or a slightly compact option
i1 <- setDT(df)[, (shift(year, type="lead", fill = year[.N])-year)==1 |
(year - shift(year, fill = year[1L]))==1, ID][, .I[.N>4 & V1] , .(rleid(V1), ID)]$V1
df[i1]
df <- data.frame(
ID=c(rep("A", 20)),
year=c(1968, 1971, 1972, 1973, 1974, 1976, 1978, 1980, 1982, 1984, 1985,
1986, 1987, 1988, 1990, 1991, 1992, 1993, 1994, 1995))
Another option using rowid
:
DT[, c("rl", "rw") := {
iscons <- cumsum(c(0L, diff(year)!=1L))
.(iscons, rowid(ID, iscons))
}]
DT[rl %in% DT[rw>=5L]$rl]
data:
#adding one more group
DT <- rbindlist(list(setDT(df), copy(df)[, ID := "B"]))
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With