Partition groups of data by group

Question

I have the following dataset:

df<- as.data.frame(c(rep("a", times = 9), rep("b", times = 18), rep("c", times = 27)))
colnames(df)<-"Location"
Year<-c(rep(1:3,times = 3), rep(1:6, times = 3), rep(1:9, times = 3))
df$Year<-Year

df<- df %>%
      mutate(Predictor = seq_along(Location)) %>%
      ungroup(df)

print(df)

Location Year Predictor
        a    1         1
        a    2         2
        a    3         3
        a    1         4
        a    2         5
        a    3         6
        a    1         7
        a    2         8
        a    3         9
        b    1        10
        b    2        11
        b    3        12
        b    4        13
        b    5        14
... 40 more rows

I want to split the above dataframe into training and test sets. For the test set, I want to randomly sample a third of the number of years in each Location, while keeping the years together. So if year "1" is selected for location "a", I want all three "1's" in the test set and so on. My test set should look something like this:

 Location Year Predictor
        a    1         1
        a    1         4
        a    1         7
        b    3        12
        b    3        18
        b    3        24
        b    5        14
        b    5        20
        b    5        26
        c    3        30
        c    3        39
        c    3        48
        c    6        33
        c    6        42
        c    6        51
        c    7        34
        c    7        43
        c    7        52

I found a similar question here, but this procedure would sample the same year and the same number of years from every location (and YEAR is numeric, not a factor). I want a different random sample of years from each location and a proportional number of samples.

Would like to do this in dplyr if possible

score 2 · Accepted Answer · answered Mar 30 '17 at 15:14

You can first create a distinct set of year/location combinations, then sample some of them for each location and use that in a semi_join on the original data. This could be done as:

df %>% 
  distinct(Location, Year) %>% 
  group_by(Location) %>% 
  sample_frac(.3) %>% 
  semi_join(df, .)

#    Location Year Predictor
# 1         a    3         3
# 2         a    3         6
# 3         a    3         9
# 4         b    4        13
# 5         b    4        19
# 6         b    4        25
# 7         b    5        14
# 8         b    5        20
# 9         b    5        26
# 10        c    8        35
# 11        c    8        44
# 12        c    8        53
# 13        c    1        28
# 14        c    1        37
# 15        c    1        46
# 16        c    2        29
# 17        c    2        38
# 18        c    2        47

Partition groups of data by group

1 Answers1