R：是否有一种干净的方法来获取循环中获得的样本的单个数据帧？

Question

I have a huge dataset containing observations about 1000 different entities. Each entity has an ID between 1 and 1000 and there are no missing IDs. Since the dataset has more than 1 million rows, I want to obtain a subset with 10 random observations for each entity to make some analysis.

The following code does the trick, but it looks cumbersome and its performance is poor.

library(dplyr) # sample_n is a dplyr function
samples <- sample_n(dataset[dataset$Entity == 1, ], 10)
for (x in 2:1000) {
  samples <- rbind(samples, sample_n(dataset[dataset$Entity == x, ], 10))
}

Could you please share some ideas for doing the same in a better fashion?

Thanks in advance!

Answer 1

I think you don't need to use a for loop when you already use dplyr . The group_by command exists to do all the work you do with your for loop in a more efficient way.

A simple example will be this:

library(dplyr)

dt = data.frame(mtcars)

dt %>% group_by(cyl) %>% sample_n(3)

To sample 3 rows for each cyl value.

So, consider that cyl here is your ID . Something like

your_dataset %>% group_by(ID) %>% sample_n(10)

will do the job.

Answer 2

As an alternative to @AntoniosKs answer why not consider using data.table now that you have a large dataset. If your data is stored as a data table in DT and you want to sample 10 observations for each ID then

library(data.table)

DT[, .SD[sample(.N,10)], by = ID]

should give you a substantial speedup.