rbindlist data.frames的列表列并选择唯一值

Question

I have a data.table 'DT' with a column ('col2') that is a list of data frames:

require(data.table)
DT <- data.table(col1 = c('A','A','B'),
                 col2 = list(data.frame(colA = c(1,3,54, 23), 
                                        colB = c("aa", "bb", "cc", "hh")),
                             data.frame(colA =c(23, 1),
                                       colB = c("hh", "aa")), 
                             data.frame(colA = 1,
                                       colB = "aa")))

> DT
   col1         col2
1:    A <data.frame>
2:    A <data.frame>
3:    B <data.frame>

>> DT$col2
[[1]]
  colA colB
1    1   aa
2    3   bb
3   54   cc
4   23   hh

[[2]]
  colA colB
1   23   hh
2    1   aa

[[3]]
  colA colB
1    1   aa

Each data.frame in col2 has two columns colA and colB. I'd like to have a data.table output that binds each unique row of those data.frames based on col1 of DT. I guess it's like using rbindlist in an aggregate function of the data.table.

This is the desired output:

> #desired output
> output
   colA colB col1
1:    1   aa    A
2:    3   bb    A
3:   54   cc    A
4:   23   hh    A
5:    1   aa    B

The dataframe of the second row of DT ( DT[2, col2] ) has duplicate entries, and only unique entries are desired for each unique col1.

I tried the following and I get an error.

desired_output <- DT[, lapply(col2, function(x) unique(rbindlist(x))), by = col1]
# Error in rbindlist(x) : 
#   Item 1 of list input is not a data.frame, data.table or list

This 'works', though not desired output:

unique(rbindlist(DT$col2))
   colA colB
1:    1   aa
2:    3   bb
3:   54   cc
4:   23   hh

Is there anyway to use rbindlist in an aggregate function of a data.table?

Answer 1

Group by 'col1', run rbindlist on 'col2':

unique(DT[ , rbindlist(col2), by = col1]) # trimmed thanks to @snoram
#    col1 colA colB
# 1:    A    1   aa
# 2:    A    3   bb
# 3:    A   54   cc
# 4:    A   23   hh
# 5:    B    1   aa

Answer 2

only unique entries are desired for each unique col1

If you add a column for col1 , the expression above means "unique entries" (unconditional on columns).

Henrik's answer is one way to keep col1 . Another is:

unique(DT[, rbindlist(setNames(col2, col1), id="col1")])

I guess this should be more efficient than

bycols = "col1"
unique(DT[, rbindlist(col2), by=bycols])   # Henrik's

though the extension to either (1) col1 not being a character column (hence suitable for setNames ) or (2) having multiple by= columns is not so obvious. For either of these cases, I would make an .id column equal to row numbers of DT then copy them over:

bycols = "col1"
res = unique(DT[, rbindlist(col2, id="DT_row")])
res[, (bycols) := DT[DT_row, ..bycols]]

To put those columns first/leftmost, I think setcolorder(res, bycols) should work, but am on too old a data.table version to see it do so.

There's also an open issue for a tidyr::unnest -like function.

Answer 3

You could do something hackish like this:

nDT <- cbind(rbindlist(DT[[2]]), col1 = rep(DT[[1]], sapply(DT[[2]], nrow)))
nDT[!duplicated(nDT)]
   colA colB col1
1:    1   aa    A
2:    3   bb    A
3:   54   cc    A
4:   23   hh    A
5:    1   aa    B

Or using tidyr (Inspired by PKumar's comment):

unique(tidyr::unnest(DT))

Or more generalisable base R:

names(DT[[2]]) <- DT[[1]]
ndf <- do.call(rbind, DT[[2]])
ndf$col1 <- substr(row.names(ndf), 1, 1)
unique(ndf)

Answer 4

This works:

DT1<-apply(DT, 1, function(x){cbind(col1=x$col1,x$col2)})
unique(rbindlist(DT1))
#   col1 colA colB
#1:    A    1   aa
#2:    A    3   bb
#3:    A   54   cc
#4:    A   23   hh
#5:    B    1   aa