CODEWITHSUNDEEP
rdplyr
Latrunculia
Latrunculia

Reputation: 726

Is there a way to sample groups of data in a grouped data frame (dplyr)

Say I have a data frame that is grouped by 2 factors. Is there a way to sample groups of data with dplyr? (note: not sample within groups)

example:

DF <- data.frame(A = rep(LETTERS[1:4], each = 6),
                 B = rep(c(1:2), 12),
                 C = rnorm(24))

# base r solution

DF$group_var <- paste(DF$A, DF$B, sep = "_")
DF_sample <- DF[DF$group_var %in% sample(unique(DF$group_var), 3), ]

#possible dplyr solution?

DF_sample <- DF %>% group_by(A,B) %>% sample_group_of_data(3)

Upvotes: 3

Views: 688

Answers (3)

ThrushJacket
ThrushJacket

Reputation: 155

I found Vincent's solution in the comments to be the one I needed. I am adding it as an additional answer.

DF %>% filter(group_var %in% sample(unique(DF$group_var), 3, replace = F))

Vincent, I owe you a +1.

Upvotes: 3

talat
talat

Reputation: 70266

Here's another pipe-solution, it works irrespective of whether the data is grouped or not:

DF %>% split(interaction(.$A, .$B)) %>% sample(3) %>% bind_rows()
# Source: local data frame [9 x 3]
# 
#       A     B          C
#   (fctr) (int)      (dbl)
# 1      B     1  0.2623781
# 2      B     1 -0.8193225
# 3      B     1  0.3348400
# 4      D     1  1.0744650
# 5      D     1  1.3528529
# 6      D     1  0.3016770
# 7      A     2 -0.1920754
# 8      A     2  0.6917583
# 9      A     2  0.1985326

The pipe itself is pretty self-explanatory, I believe.

Upvotes: 4

Choubi
Choubi

Reputation: 680

Probably not as pretty as you would have wanted and it's kind of cheating but, here's my solution:

DF %>% group_by(A, B) %>%
  magrittr::extract(unlist(sample(attr(., "indices"), 5))+1, )

I use the "indices" attribute which gives the indices of the groups in a list. I sample this list, unlist it and add 1 (it seems those indices start at 0).

I then use the magrittr extract function which stands for the [] operator. In that sense I am kind of cheating as effectively, I have just rewritten with the pipe your problem and using the attributes from the grouped dplyr.

Upvotes: 2

Related Questions