Add `n_unique()` = length(unique(x)) - useful while grouping

[arunsrinivasan]

The SO question that triggered the thought:

Instead of having to do:

DT[, length(unique(.)), by=.]

We could do with:

DT[, n_unique(.), by=.]

This'll especially be faster for data.tables though because we don't have to subset the entire data.table to know the number of unique values.

Here's a quick benchmark:

require(data.table)
x = sample(1e2, 1e7, TRUE)
system.time(ans1 <- length(unique(x))) # 0.667 seconds
system.time(ans2 <- length(attr(data.table:::forderv(x, retGrp=TRUE), 'starts'))) # 0.1 seconds

We could, in addition, also internally optimise length(unique(.)) to n_unique(.).

[matthieugomez]

I'd like this.

It would be nice if the function also had a method for data.table: relevant SO question here and there. For instance to find the number of unique combinations (v2, v3) within groups defined by v1, one could do:

DT[, n_unique(.SD), by = v1, .SDcols = c("v2", "v3")]

[arunsrinivasan]

Implemented as uniqueN.

[matthieugomez]

It would be nice if the syntax was made closer to unique with a by option (if by is not given, default to keys)

[jangorecki]

@matthieugomez isn't that already incorporated?
https://github.com/Rdatatable/data.table/blob/master/R/duplicated.R#L93

[matthieugomez]

Great! Sorry I was looking at the original commit.