Using gather() to gather two (or more) groups of columns into two (or more) key-value pairs [duplicate]

Question

This question already has an answer here:

• Reshaping multiple sets of measurement columns (wide format) into single columns (long format) 7 answers

I want to gather two seperate groups of columns into two key-value pairs. Here's some example data:

library(dplyr)
library(tidyr)
ID = c(1:5)
measure1 = c(1:5)
measure2 = c(6:10)
letter1 = c("a", "b", "c", "d", "e")
letter2 = c("f", "g", "h", "i", "j")

df = data.frame(ID, measure1, measure2, letter1, letter2)
df = tbl_df(df)
df$letter1 <- as.character(df$letter1)
df$letter2 <- as.character(df$letter2)

I want the values of the two measure columns (measure1 and measure2) to be in one column with a key-column next to it (the key-value pair). I also want the same for letter1 and letter2. I figured that I could use select() to create two different datasets, use gather seperately on both datasets and then join (this worked):

df_measure = df %>% 
  select(ID, measure1, measure2) %>% 
  gather(measure_time, measure, -ID) %>% 
  mutate(id.extra = c(1:10))
df_letter = df %>% 
  select(ID, letter1, letter2) %>% 
  gather(letter_time, letter, -ID) %>% 
  mutate(id.extra = c(1:10))
df_long = df_measure %>% 
  left_join(df_letter, by = "id.extra")

So this works perfectly (in this case), but i guess this could be done more elegantly (without stuff like splitting or creating 'id.extra').So please shed some light on it!

Answer 1

You can use something like the following. I'm not sure from your current approach if this is exactly your desired output or not since it seems to contain a lot of redundant information.

df %>%
  gather(val, var, -ID) %>%
  extract(val, c("value", "time"), regex = "([a-z]+)([0-9]+)") %>%
  spread(value, var)
# # A tibble: 10 × 4
#       ID  time letter measure
# *  <int> <chr>  <chr>   <chr>
# 1      1     1      a       1
# 2      1     2      f       6
# 3      2     1      b       2
# 4      2     2      g       7
# 5      3     1      c       3
# 6      3     2      h       8
# 7      4     1      d       4
# 8      4     2      i       9
# 9      5     1      e       5
# 10     5     2      j      10

---

This is much more easily done with melt + patterns from "data.table":

library(data.table)
melt(as.data.table(df), measure.vars = patterns("measure", "letter"))

---

Or you can be old-school and just use reshape from base R. Note, however, that base R's reshape does not like "tibbles", so you have to convert it with as.data.frame).

reshape(as.data.frame(df), direction = "long", idvar = "ID", 
        varying = 2:ncol(df), sep = "")

Answer 2

We can use melt from data.table which can take multiple measure patterns

library(data.table)
melt(setDT(df), measure = patterns("^measure", "^letter"), 
          value.name = c("measure", "letter"))
#     ID variable measure letter
# 1:  1        1       1      a
# 2:  2        1       2      b
# 3:  3        1       3      c
# 4:  4        1       4      d
# 5:  5        1       5      e
# 6:  1        2       6      f
# 7:  2        2       7      g
# 8:  3        2       8      h
# 9:  4        2       9      i
#10:  5        2      10      j