R: Unique count by first occurrence of grouping variable - r

Question

I would like to create a new variable "Count" that is a count of the unique values of a factor "Period", by grouping variable "ID". The following data includes a column with the values I would want in "Count":
structure(list(ID = structure(c(1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L
), .Label = c("a", "b"), class = "factor"), Period = c(1.1, 1.1,
1.2, 1.3, 1.2, 1.3, 1.5, 1.5), Count = c(1L, 1L, 2L, 3L, 1L,
2L, 3L, 3L)), .Names = c("ID", "Period", "Count"), class = "data.frame", row.names = c(NA,
-8L))
I tried to use mutate with Count = 1:length(Period) but it creates a cumulative count of each value of "Period", whereas I want a cumulative count of only unique values. This is what I tried:
library(plyr)
samp1<-ddply(samp, .(ID, Period), mutate, Count = 1:length(Period))
Could anyone provide the correct function to use?

Answer 1

Edit- New answer
Now that come to think of it some more, my initial approach won't return correct results if each groups elements aren't grouped together, so for example for
v <- c(1, 3, 2, 2, 1, 2)
My function will put non-consecutive 1s and 2 in different groups
myrleid(v)
## [1] 1 2 3 3 4 5
Thus, the best approach seem to be
match(v, unique(v))
## [1] 1 2 3 3 1 3
Will will both preserve the appearance order and keep un-ordered values in the same group.
Thus, I would recommend just doing
library(data.table)
setDT(df)[, Count2 := match(Period, unique(Period)), by = ID]
or (with base R)
with(df, ave(Period, ID, FUN = function(x) match(x, unique(x))))
Old answer
Looks like a good candidate for the rleid function from the data.table devel version on GH
### Devel version installation instructions
# library(devtools)
# install_github("Rdatatable/data.table", build_vignettes = FALSE)
library(data.table) # v 1.9.5+
setDT(df)[, Count2 := rleid(Period), by = ID]
df
# ID Period Count Count2
# 1: a 1.1 1 1
# 2: a 1.1 1 1
# 3: a 1.2 2 2
# 4: a 1.3 3 3
# 5: b 1.2 1 1
# 6: b 1.3 2 2
# 7: b 1.5 3 3
# 8: b 1.5 3 3
Or, If you don't want to load external packages, we could define this function on our own
myrleid <- function(x) {
temp <- rle(x)$lengths
rep.int(seq_along(temp), temp)
}
with(df, ave(Period, ID, FUN = myrleid))
## [1] 1 1 2 3 1 2 3 3
Or if the groups are in increasing order, you could try ranking them too
library(data.table) ## V1.9.5+
setDT(df)[, Count2 := frank(Period, ties.method = "dense"), by = ID]
Or
library(dplyr)
df %>%
group_by(ID) %>%
mutate(Count2 = dense_rank(Period))

Answer 2

samp <- structure(list(ID = structure(c(1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L
), .Label = c("a", "b"), class = "factor"), Period = c(1.1, 1.1,
1.2, 1.3, 1.2, 1.3, 1.5, 1.5), Count = c(1L, 1L, 2L, 3L, 1L,
2L, 3L, 3L)), .Names = c("ID", "Period", "Count"), class = "data.frame", row.names = c(NA,
-8L))
select(samp, -Count) %>%
arrange(ID, Period) %>%
group_by(ID) %>%
mutate(dup = !duplicated(Period),
Count = cumsum(dup))
The key steps are to arrange by ID and Period, and then to identify that first new representation of Period as "not duplicated".

Answer 3

A solution in base R with transform:
transform(df, Count2 = unlist(
tapply(df$Period, df$ID, function(x)
as.numeric(factor(x)))
))
ID Period Count Count2
a1 a 1.1 1 1
a2 a 1.1 1 1
a3 a 1.2 2 2
a4 a 1.3 3 3
b1 b 1.2 1 1
b2 b 1.3 2 2
b3 b 1.5 3 3
b4 b 1.5 3 3
as David suggested this solution does not work well if data Period are not monotonic increasing.