Replace values in a data.frame column based on values in a different column

Question

I have this data.frame:

df <- data.frame(id = rep(c("one", "two", "three"), each = 10), week.born = NA)
df$week.born[c(5,15,28)] <- c(23,19,24)

df 

  id week.born
1    one        NA
2    one        NA
3    one        NA
4    one        NA
5    one        23
6    one        NA
7    one        NA
8    one        NA
9    one        NA
10   one        NA
11   two        NA
12   two        NA
13   two        NA
14   two        NA
15   two        19
16   two        NA
17   two        NA
18   two        NA
19   two        NA
20   two        NA
21 three        NA
22 three        NA
23 three        NA
24 three        NA
25 three        NA
26 three        NA
27 three        NA
28 three        24
29 three        NA
30 three        NA

For one all week.born values should be 23. For two all week.born values should be 19. For one all week.born values should be 24.

Whats the best way to do this?

what if there are two or more non NA values in week.born column for a single id? — Colonel Beauvel
– Colonel Beauvel, Commented Apr 22, 2015 at 19:33
possible duplicate of Replace values in data frame with other values according to a rule — Sam Firke
– Sam Firke, Commented Apr 22, 2015 at 19:39

David Arenburg · Accepted Answer · 2015-04-22 21:49:22Z

6

I would create another data.frame containing the mapping and then do a simple join:

require(dplyr)
map <- data.frame(id=c("one","two","three"), new.week.born=c(23,19,24))
left_join(df, map, by="id")

# id week.born new.week.born
# 1    one        NA            23
# 2    one        NA            23
# ...
# 16   two        NA            19
# 17   two        NA            19
# 18   two        NA            19
# 19   two        NA            19
# 20   two        NA            19
# 21 three        NA            24
# 22 three        NA            24
# 23 three        NA            24
# ...

See benchmark below.

library(microbenchmark)
library(dplyr) # v 0.4.1
library(data.table) # v 1.9.5

df <- data.frame(id = rep(c("one", "two", "three"), each = 1e6))
df2 <- copy(df)
map <- data.frame(id=c("one","two","three"), new.week.born=c(23,19,24))

dplyr_join <- function() { 
  left_join(df, map, by="id")
}

r_merge <- function() {
  merge(df, map, by="id")
}

data.table_join <- function() {
  setkey(setDT(df2))[map]
}

Unit: milliseconds
              expr         min         lq       mean     median         uq       max neval
      dplyr_join()   409.10635   476.6690   910.6446   489.4573   705.4021  2866.151    10
         r_merge() 41589.32357 47376.0741 55719.1752 50133.0918 54636.3356 83562.931    10
 data.table_join()    94.14621   132.3788   483.4220   225.3309  1051.7916  1416.946    10

edited Apr 22, 2015 at 21:49

David Arenburg

92.4k18 gold badges145 silver badges202 bronze badges

answered Apr 22, 2015 at 19:33

faidherbard

1,0359 silver badges20 bronze badges

Sign up to request clarification or add additional context in comments.

8 Comments

David Arenburg Over a year ago

You did a bit strange benchmark I might say. Why did you recreate each data set in each function? Also, Jan solution can be shorten to just setkey(setDT(df))[lkp]. Finally, what data.table version do you have?

faidherbard Over a year ago

Made the comparison with dplyr_0.4.1 and data.table_1.9.4. I've re-created the dataset in each function to avoid having it modified by the data.table package.

David Arenburg Over a year ago

My benchmark on data.table v 1.9.5 came much faster than dplyrs. But you need shorten the code.

faidherbard Over a year ago

Ok for setkey(setDT(df))[lkp] however it doesn't improve the performances and it makes is even more tricky to understand the code. That was my point with dplyr which keeps it quite simple with the left_join syntax and which doesn't put any prequisite on the data.frame (no key required)

Arun Over a year ago

faidherbard, your point about joins has been extensively covered in this data.table vs dplyr post.

|

Colonel Beauvel · Accepted Answer · 2015-04-22 19:35:34Z

3

You can do:

library(data.table)
setDT(df)[,week.born:=week.born[!is.na(week.born)][1], by=id]

Or base R using ave:

df$week.born = with(df, ave(week.born, id, FUN=function(u) u[!is.na(u)][1]))

answered Apr 22, 2015 at 19:35

Colonel Beauvel

31.3k11 gold badges49 silver badges88 bronze badges

Comments

cho7tom · Accepted Answer · 2015-04-22 19:30:46Z

2

One solution is:

df$week.born[df$id == "one"] <- 23
df$week.born[df$id == "two"] <- 19
df$week.born[df$id == "three"] <- 24

Regards

answered Apr 22, 2015 at 19:30

cho7tom

1,0702 gold badges13 silver badges30 bronze badges

Comments

David Arenburg · Accepted Answer · 2015-04-22 20:51:39Z

2

@cho7tom is OK if you have just few groups, otherwise you may prefer to have a lookup table and make a join to that table to lookup week.born value based on id.

base R

df <- data.frame(id = rep(c("one", "two", "three"), each = 10))
lkp <- data.frame(id=c("one","two","three"), week.born=c(23,19,24))
merge(df, lkp, by="id")

Or using binary join from data.table

library(data.table)
setkey(setDT(df))[lkp]

edited Apr 22, 2015 at 20:51

David Arenburg

92.4k18 gold badges145 silver badges202 bronze badges

answered Apr 22, 2015 at 19:37

jangorecki

16.8k5 gold badges86 silver badges169 bronze badges

Comments

Sam Firke · Accepted Answer · 2015-04-23 12:34:27Z

0

When mapping just a few combinations like this, the mapvalues function from the plyr package is simple:

library(plyr)
df$week.born <- mapvalues(df$id, c("one", "two", "three"), c(23, 19, 24))

answered Apr 23, 2015 at 12:34

Sam Firke

23.4k11 gold badges100 silver badges117 bronze badges

Collectives™ on Stack Overflow

Replace values in a data.frame column based on values in a different column

5 Answers 5

8 Comments

Comments

Comments

Comments

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

5 Answers 5

8 Comments

Comments

Comments

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related