data.table: Sum by all existing combinations in table

Question

I have a data.table out like this (in reality it is much larger):

out <-      code weights group
        1:    2   0.387      1
        2:            


        

Answer 1

Using CJ (cross join) you can add the missing combinations:

library(data.table)
setkey(out, code, group)
out[CJ(code, group, unique = TRUE)
    ][, lapply(.SD, sum), by = .(code, group)
      ][is.na(weights), weights := 0]

gives:

   code group weights
1:    1     1   0.399
2:    1     2   0.212
3:    1     3   0.474
4:    2     1   1.997
5:    2     2   0.373
6:    2     3   0.569
7:    3     1   0.000
8:    3     2   1.323
9:    3     3   0.316

---

Or with xtabs as @alexis_laz showed in the comments:

xtabs(weights ~ group + code, out)

which gives:

     code
group     1     2     3
    1 0.399 1.997 0.000
    2 0.212 0.373 1.323
    3 0.474 0.569 0.316

If you want to get this output in a long-form dataframe, you can wrap the xtabs code in the melt function of the reshape2 (or data.table) package:

library(reshape2)
res <- melt(xtabs(weights ~ group + code, out))

which gives:

> class(res)
[1] "data.frame"
> res
  group code value
1     1    1 0.399
2     2    1 0.212
3     3    1 0.474
4     1    2 1.997
5     2    2 0.373
6     3    2 0.569
7     1    3 0.000
8     2    3 1.323
9     3    3 0.316

---

You could also do this with a combination of dplyr and tidyr:

library(dplyr)
library(tidyr)
out %>%
  complete(code, group, fill = list(weights=0)) %>%
  group_by(code, group) %>% 
  summarise(sum(weights))