Repeat rows of a data.frame

I want to repeat the rows of a data.frame, each N times. The result should be a new data.frame (with nrow(new.df) == nrow(old.df) * N) keeping the data types of the columns.

Example for N = 2:

                        A B   C
  A B   C             1 j i 100
1 j i 100     -->     2 j i 100
2 K P 101             3 K P 101
                      4 K P 101

So, each row is repeated 2 times and characters remain characters, factors remain factors, numerics remain numerics, ...

My first attempt used apply: apply(old.df, 2, function(co) rep(co, each = N)), but this one transforms my values to characters and I get:

     A   B   C    
[1,] "j" "i" "100"
[2,] "j" "i" "100"
[3,] "K" "P" "101"
[4,] "K" "P" "101"

df <- data.frame(a = 1:2, b = letters[1:2]) 
df[rep(seq_len(nrow(df)), each = 2), ]

A clean dplyr solution, taken from [here][1]

library(dplyr)
df <- tibble(x = 1:2, y = c("a", "b"))
df %>% slice(rep(1:n(), each = 2))

[1]: https://stackoverflow.com/questions/38237350/repeating-rows-of-data-frame-in-dplyr#comment63896747_38237805 "here"

There is a lovely vectorized solution that repeats only certain rows n-times each, possible for example by adding an ntimes column to your data frame:

  A B   C ntimes
1 j i 100      2
2 K P 101      4
3 Z Z 102      1

Method:

df <- data.frame(A=c("j","K","Z"), B=c("i","P","Z"), C=c(100,101,102), ntimes=c(2,4,1))
df <- as.data.frame(lapply(df, rep, df$ntimes))

Result:

  A B   C ntimes
1 Z Z 102      1
2 j i 100      2
3 j i 100      2
4 K P 101      4
5 K P 101      4
6 K P 101      4
7 K P 101      4

This is very similar to Josh O'Brien and Mark Miller's method:

df[rep(seq_len(nrow(df)), df$ntimes),]

However, that method appears quite a bit slower:

df <- data.frame(A=c("j","K","Z"), B=c("i","P","Z"), C=c(100,101,102), ntimes=c(2000,3000,4000))

microbenchmark::microbenchmark(
  df[rep(seq_len(nrow(df)), df$ntimes),],
  as.data.frame(lapply(df, rep, df$ntimes)),
  times = 10
)

Result:

Unit: microseconds
                                      expr      min       lq      mean   median       uq      max neval
   df[rep(seq_len(nrow(df)), df$ntimes), ] 3563.113 3586.873 3683.7790 3613.702 3657.063 4326.757    10
 as.data.frame(lapply(df, rep, df$ntimes))  625.552  654.638  676.4067  668.094  681.929  799.893    10

If you can repeat the whole thing, or subset it first then repeat that, then this similar question may be helpful. Once again:

library(mefa)
rep(mtcars,10) 

or simply

mefa:::rep.data.frame(mtcars)

Adding to what @dardisco mentioned about mefa::rep.data.frame(), it's very flexible.

You can either repeat each row N times:

rep(df, each=N)

or repeat the entire dataframe N times (think: like when you recycle a vectorized argument)

rep(df, times=N)

Two thumbs up for mefa! I had never heard of it until now and I had to write manual code to do this.

For reference and adding to answers citing mefa, it might worth to take a look on the implementation of mefa::rep.data.frame() in case you don't want to include the whole package:

> data <- data.frame(a=letters[1:3], b=letters[4:6])
> data
  a b
1 a d
2 b e
3 c f
> as.data.frame(lapply(data, rep, 2))
  a b
1 a d
2 b e
3 c f
4 a d
5 b e
6 c f

The rep.row function seems to sometimes make lists for columns, which leads to bad memory hijinks. I have written the following which seems to work well:

library(plyr)
rep.row <- function(r, n){
  colwise(function(x) rep(x, n))(r)
}

My solution similar as mefa:::rep.data.frame, but a little faster and cares about row names:

rep.data.frame <- function(x, times) {
    rnames <- attr(x, "row.names")
    x <- lapply(x, rep.int, times = times)
    class(x) <- "data.frame"
    if (!is.numeric(rnames))
        attr(x, "row.names") <- make.unique(rep.int(rnames, times))
    else
        attr(x, "row.names") <- .set_row_names(length(rnames) * times)
    x
}

Compare solutions:

library(Lahman)
library(microbenchmark)
microbenchmark(
    mefa:::rep.data.frame(Batting, 10),
    rep.data.frame(Batting, 10),
    Batting[rep.int(seq_len(nrow(Batting)), 10), ],
    times = 10
)
#> Unit: milliseconds
#>                                            expr       min       lq     mean   median        uq       max neval cld
#>              mefa:::rep.data.frame(Batting, 10) 127.77786 135.3480 198.0240 148.1749  278.1066  356.3210    10  a 
#>                     rep.data.frame(Batting, 10)  79.70335  82.8165 134.0974  87.2587  191.1713  307.4567    10  a 
#>  Batting[rep.int(seq_len(nrow(Batting)), 10), ] 895.73750 922.7059 981.8891 956.3463 1018.2411 1127.3927    10   b

try using for example

N=2
rep(1:4, each = N) 

as an index

Another way to do this would to first get row indices, append extra copies of the df, and then order by the indices:

df$index = 1:nrow(df)
df = rbind(df,df)
df = df[order(df$index),][,-ncol(df)]

Although the other solutions may be shorter, this method may be more advantageous in certain situations.