寻找一种基于多变量组内的值差异来过滤data.table的更简洁方法

Question

I was working on a problem where I had two grouping variables and one value. I only to keep the rows were at least two of the values in the group are close to each other in value. In the example I wanted groups that had one set of values within 10 of each other.

Below is what I initially tried, and something about making a flag variable made me feel like I was doing it in some roundabout way, and I just wanted to know if there's a cleaner more intended way to do something like this in data.table. Thank you!

x and y are the categories, z is the value.

library(data.table)
set.seed(123)


dt <- data.table(
  x = sample(LETTERS, 1000, T),
  y = sample(letters, 1000, T),
  z = sample(100, 1000, T),
  key = tail(letters, 3)
)

dt <- unique(dt)
dt <- dt[dt[, .(flag = any(diff(z) <= 11)), .(x, y)], on = c("x", "y")][(flag)]
dt[, flag := NULL]
dt

Answer 1

You can use .I with an if to determine whether to include each group (here want matches your final dt )

dt <- unique(dt)
want <- dt[dt[, if(any(diff(z) <= 11)) .I, .(x, y)]$V1]

Answer 2

You could do

res <- dt[, if (.N > 1L && min(diff(z)) <= 11) .SD, by=.(x, y)]

I used min instead of any since I guess it leads to fewer computations.

I added the .N > 1L condition since you need to think about how to handle the single row case (where diff is NA). You could do

• .N > 1L && to drop those cases
• .N == 1L || to keep them

I just wanted to know if there's a cleaner more intended way to do something like this in data.table

I think a having= syntax would be convenient for this. It's currently a feature request .

---

Input data (since OP overwrites it):

library(data.table)
set.seed(123)
dt <- data.table(
  x = sample(LETTERS, 1000, T),
  y = sample(letters, 1000, T),
  z = sample(100, 1000, T),
  key = tail(letters, 3)
)
dt <- unique(dt)