9.3 Random Permutation and Random Sampling

Now, you have covered how to work with distributions in R with the four useful functions for each distribution. In many applications, you may want to randomly permute or sample elements from a vector. Let’s see how to do that. The vector x <- 6:10 will be used throughout this section.

9.3.1 Random Permutation

In statistics and machine learning, you usually need to do a random permutation of the data. For example, you can evaluate a model’s performance by dividing the data randomly into two parts for training and validation, respectively.

For the vector x <- 6:10, you can use the function sample() to get a permutation for x.

x <- 6:10
set.seed(0)
sample(x)  #a random permutation of x
#> [1]  6  9  8 10  7

To reproduce the random permutation, we can use the same seed.

set.seed(0)
sample(x)  #reproduce the random permutation
#> [1]  6  9  8 10  7

9.3.2 Random Sampling without Replacement

Note that the vector x has 5 elements in total. To sample a few elements from x, you can again use the sample() function. For example, if you want to randomly sample two elements from x, you can use the following code

sample(x, size = 2) 
#> [1] 10  8

Here, the size argument specify the targeted number of elements. By default, the sample function take a sample without replacement, i.e. the results sample has no duplicated elements. Because of this, if the size is larger than the length of the vector x, you will see an error message as follows.

sample(x, size = 6)
#> Error in sample.int(length(x), size, replace, prob): cannot take a sample larger than the population when 'replace = FALSE'

In addition to using a vector in the first argument of sample, you can also use a positive integer (e.g., 10), which will be equivalent to x = 1:10. See the following code for an example.

sample(10, size = 4)   #sample 4 integers from 1 to 10.
#> [1] 2 3 1 5
sample(1:10, size = 4) #sample 4 integers from 1 to 10.
#> [1] 5 6 9 2

9.3.3 Random Sampling with Replacement

Sometimes, you may want to get a sample with replacements. You will still be using the sample function, but setting the argument replace = TRUE. The following code samples 10 elements with replacement from x.

sample(x, size = 10, replace = TRUE)
#>  [1]  6 10 10  6  6 10 10  7  7  6

As expected, you will see some duplicated elements in the output vector.

A very important application of random sample with replacement is bootstrap. A bootstrap sample is a sample of the same size as the original data with replacement. So, if you want to get a bootstrap sample from x, you will sample 5 elements with replacement from x.

sample(x, replace = TRUE) #a bootstrap sample
#> [1] 9 6 9 8 7

Note that, when the argument size is not provided, it will take the default value: the length of x.

9.3.4 Random Sampling with Unequal Probabilities

By default, the sample() function will draw each element with the same probability. In some cases, you may want to assign different probabilities for different elements.

To draw elements with different probabilities, the first method is to use the random number generator for Binomial distribution or Bernoulli distribution. Let’s say we want to randomly sample 100 elements from a Bernoulli distribution with success probability \(p=0.2\).

rbinom(100, size = 1, prob = 0.2)
#>   [1] 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 1 0 0 0 1 1
#>  [38] 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 0 1 0 0 0
#>  [75] 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0

In addition to using the rbinom function introduced in Section 9.2, you can use the sample function with the prob argument inside to achieve the same goal.

sample(c(0, 1), size = 100, replace = TRUE, prob = c(0.8, 0.2))
#>   [1] 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 1 1 0 0 1 0
#>  [38] 0 0 1 0 0 1 0 1 0 0 0 1 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1
#>  [75] 1 0 0 1 1 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0

You will samples 100 elements with replacement from c(0,1) here, and the probability of drawing 0 is 0.8, the probability of drawing 1 is 0.2.