Split data frame string column into multiple columns

Question

I'd like to take data of the form

before = data.frame(attr = c(1,30,4,6), type=c('foo_and_bar','foo_and_bar_2'))
  attr          type
1    1   foo_and_bar
2   30 foo_and_bar_2
3    4   foo_and_bar
4    6 foo_and_bar_2

and use split() on the column "type" from above to get something like this:

  attr type_1 type_2
1    1    foo    bar
2   30    foo  bar_2
3    4    foo    bar
4    6    foo  bar_2

I came up with something unbelievably complex involving some form of apply that worked, but I've since misplaced that. It seemed far too complicated to be the best way. I can use strsplit as below, but then unclear how to get that back into 2 columns in the data frame.

> strsplit(as.character(before$type),'_and_')
[[1]]
[1] "foo" "bar"

[[2]]
[1] "foo"   "bar_2"

[[3]]
[1] "foo" "bar"

[[4]]
[1] "foo"   "bar_2"

Thanks for any pointers. I've not quite groked R lists just yet.

Answer 1

Use stringr::str_split_fixed

library(stringr)
str_split_fixed(before$type, "_and_", 2)

Answer 2

Another option is to use the new tidyr package.

library(dplyr)
library(tidyr)

before <- data.frame(
  attr = c(1, 30 ,4 ,6 ), 
  type = c('foo_and_bar', 'foo_and_bar_2')
)

before %>%
  separate(type, c("foo", "bar"), "_and_")

##   attr foo   bar
## 1    1 foo   bar
## 2   30 foo bar_2
## 3    4 foo   bar
## 4    6 foo bar_2

Answer 3

5 years later adding the obligatory data.table solution

library(data.table) ## v 1.9.6+ 
setDT(before)[, paste0("type", 1:2) := tstrsplit(type, "_and_")]
before
#    attr          type type1 type2
# 1:    1   foo_and_bar   foo   bar
# 2:   30 foo_and_bar_2   foo bar_2
# 3:    4   foo_and_bar   foo   bar
# 4:    6 foo_and_bar_2   foo bar_2

We could also both make sure that the resulting columns will have correct types and improve performance by adding type.convert and fixed arguments (since "_and_" isn't really a regex)

setDT(before)[, paste0("type", 1:2) := tstrsplit(type, "_and_", type.convert = TRUE, fixed = TRUE)]

Answer 4

Yet another approach: use rbind on out:

before <- data.frame(attr = c(1,30,4,6), type=c('foo_and_bar','foo_and_bar_2'))  
out <- strsplit(as.character(before$type),'_and_') 
do.call(rbind, out)

     [,1]  [,2]   
[1,] "foo" "bar"  
[2,] "foo" "bar_2"
[3,] "foo" "bar"  
[4,] "foo" "bar_2"

And to combine:

data.frame(before$attr, do.call(rbind, out))

Answer 5

Notice that sapply with "[" can be used to extract either the first or second items in those lists so:

before$type_1 <- sapply(strsplit(as.character(before$type),'_and_'), "[", 1)
before$type_2 <- sapply(strsplit(as.character(before$type),'_and_'), "[", 2)
before$type <- NULL

And here's a gsub method:

before$type_1 <- gsub("_and_.+$", "", before$type)
before$type_2 <- gsub("^.+_and_", "", before$type)
before$type <- NULL

Answer 6

here is a one liner along the same lines as aniko's solution, but using hadley's stringr package:

do.call(rbind, str_split(before$type, '_and_'))

Answer 7

To add to the options, you could also use my splitstackshape::cSplit function like this:

library(splitstackshape)
cSplit(before, "type", "_and_")
#    attr type_1 type_2
# 1:    1    foo    bar
# 2:   30    foo  bar_2
# 3:    4    foo    bar
# 4:    6    foo  bar_2

Answer 8

The subject is almost exhausted, I 'd like though to offer a solution to a slightly more general version where you don't know the number of output columns, a priori. So for example you have

before = data.frame(attr = c(1,30,4,6), type=c('foo_and_bar','foo_and_bar_2', 'foo_and_bar_2_and_bar_3', 'foo_and_bar'))
  attr                    type
1    1             foo_and_bar
2   30           foo_and_bar_2
3    4 foo_and_bar_2_and_bar_3
4    6             foo_and_bar

We can't use dplyr separate() because we don't know the number of the result columns before the split, so I have then created a function that uses stringr to split a column, given the pattern and a name prefix for the generated columns. I hope the coding patterns used, are correct.

split_into_multiple <- function(column, pattern = ", ", into_prefix){
  cols <- str_split_fixed(column, pattern, n = Inf)
  # Sub out the ""'s returned by filling the matrix to the right, with NAs which are useful
  cols[which(cols == "")] <- NA
  cols <- as.tibble(cols)
  # name the 'cols' tibble as 'into_prefix_1', 'into_prefix_2', ..., 'into_prefix_m' 
  # where m = # columns of 'cols'
  m <- dim(cols)[2]

  names(cols) <- paste(into_prefix, 1:m, sep = "_")
  return(cols)
}

We can then use split_into_multiple in a dplyr pipe as follows:

after <- before %>% 
  bind_cols(split_into_multiple(.$type, "_and_", "type")) %>% 
  # selecting those that start with 'type_' will remove the original 'type' column
  select(attr, starts_with("type_"))

>after
  attr type_1 type_2 type_3
1    1    foo    bar   <NA>
2   30    foo  bar_2   <NA>
3    4    foo  bar_2  bar_3
4    6    foo    bar   <NA>

And then we can use gather to tidy up...

after %>% 
  gather(key, val, -attr, na.rm = T)

   attr    key   val
1     1 type_1   foo
2    30 type_1   foo
3     4 type_1   foo
4     6 type_1   foo
5     1 type_2   bar
6    30 type_2 bar_2
7     4 type_2 bar_2
8     6 type_2   bar
11    4 type_3 bar_3

Answer 9

An easy way is to use sapply() and the [ function:

before <- data.frame(attr = c(1,30,4,6), type=c('foo_and_bar','foo_and_bar_2'))
out <- strsplit(as.character(before$type),'_and_')

For example:

> data.frame(t(sapply(out, `[`)))
   X1    X2
1 foo   bar
2 foo bar_2
3 foo   bar
4 foo bar_2

sapply()'s result is a matrix and needs transposing and casting back to a data frame. It is then some simple manipulations that yield the result you wanted:

after <- with(before, data.frame(attr = attr))
after <- cbind(after, data.frame(t(sapply(out, `[`))))
names(after)[2:3] <- paste("type", 1:2, sep = "_")

At this point, after is what you wanted

> after
  attr type_1 type_2
1    1    foo    bar
2   30    foo  bar_2
3    4    foo    bar
4    6    foo  bar_2

Answer 10

Since R version 3.4.0 you can use strcapture() from the utils package (included with base R installs), binding the output onto the other column(s).

out <- strcapture(
    "(.*)_and_(.*)",
    as.character(before$type),
    data.frame(type_1 = character(), type_2 = character())
)

cbind(before["attr"], out)
#   attr type_1 type_2
# 1    1    foo    bar
# 2   30    foo  bar_2
# 3    4    foo    bar
# 4    6    foo  bar_2

Answer 11

Here is a base R one liner that overlaps a number of previous solutions, but returns a data.frame with the proper names.

out <- setNames(data.frame(before$attr,
                  do.call(rbind, strsplit(as.character(before$type),
                                          split="_and_"))),
                  c("attr", paste0("type_", 1:2)))
out
  attr type_1 type_2
1    1    foo    bar
2   30    foo  bar_2
3    4    foo    bar
4    6    foo  bar_2

It uses strsplit to break up the variable, and data.frame with do.call/rbind to put the data back into a data.frame. The additional incremental improvement is the use of setNames to add variable names to the data.frame.

Answer 12

This question is pretty old but I'll add the solution I found the be the simplest at present.

library(reshape2)
before = data.frame(attr = c(1,30,4,6), type=c('foo_and_bar','foo_and_bar_2'))
newColNames <- c("type1", "type2")
newCols <- colsplit(before$type, "_and_", newColNames)
after <- cbind(before, newCols)
after$type <- NULL
after

Answer 13

base but probably slow:

n <- 1
for(i in strsplit(as.character(before$type),'_and_')){
     before[n, 'type_1'] <- i[[1]]
     before[n, 'type_2'] <- i[[2]]
     n <- n + 1
}

##   attr          type type_1 type_2
## 1    1   foo_and_bar    foo    bar
## 2   30 foo_and_bar_2    foo  bar_2
## 3    4   foo_and_bar    foo    bar
## 4    6 foo_and_bar_2    foo  bar_2

Answer 14

Another approach if you want to stick with strsplit() is to use the unlist() command. Here's a solution along those lines.

tmp <- matrix(unlist(strsplit(as.character(before$type), '_and_')), ncol=2,
   byrow=TRUE)
after <- cbind(before$attr, as.data.frame(tmp))
names(after) <- c("attr", "type_1", "type_2")

Answer 15

Here is another base R solution. We can use read.table but since it accepts only one-byte sep argument and here we have multi-byte separator we can use gsub to replace the multibyte separator to any one-byte separator and use that as sep argument in read.table

cbind(before[1], read.table(text = gsub('_and_', '\t', before$type), 
                 sep = "\t", col.names = paste0("type_", 1:2)))

#  attr type_1 type_2
#1    1    foo    bar
#2   30    foo  bar_2
#3    4    foo    bar
#4    6    foo  bar_2

In this case, we can also make it shorter by replacing it with default sep argument so we don't have to mention it explicitly

cbind(before[1], read.table(text = gsub('_and_', ' ', before$type), 
                 col.names = paste0("type_", 1:2)))

Answer 16

Surprisingly, another tidyverse solution is still missing - you can also use tidyr::extract, with a regex.

library(tidyr)
before <- data.frame(attr = c(1, 30, 4, 6), type = c("foo_and_bar", "foo_and_bar_2"))

## regex - getting all characters except an underscore till the first underscore, 
## inspired by Akrun https://stackoverflow.com/a/49752920/7941188 

extract(before, col = type, into = paste0("type", 1:2), regex = "(^[^_]*)_(.*)")
#>   attr type1     type2
#> 1    1   foo   and_bar
#> 2   30   foo and_bar_2
#> 3    4   foo   and_bar
#> 4    6   foo and_bar_2

Answer 17

Another base R solution that also is a general way to split a column in several columns is:

Data

before = data.frame(attr = c(1,30,4,6), type=c('foo_and_bar','foo_and_bar_2'))

Procedure

attach(before)
before$type2 <- gsub("(\\w*)_and_(\\w*)", "c('\\1', '\\2')", type)
#this recode the column type to c("blah", "blah") form

cbind(before,t(sapply(1:nrow(before), function(x) eval(parse(text=before$type2[x])))))
#this split the desired column into several ones named 1 2 3 and so on

OUTPUT

  attr          type             type2   1     2
1    1   foo_and_bar   c('foo', 'bar') foo   bar
2   30 foo_and_bar_2 c('foo', 'bar_2') foo bar_2
3    4   foo_and_bar   c('foo', 'bar') foo   bar
4    6 foo_and_bar_2 c('foo', 'bar_2') foo bar_2

Answer 18

Since this question was asked separate has been superseded by separate_longer_* and separate_wider_* functions.

The way to do it now is:

library(tidyr)
separate_wider_delim(before, type, delim = "_and_", names_sep = "_")

You could also use separate_wider_regex, but I'll leave that as an exercise to the reader :-)