tests/testthat/test-utilities-noise-generation.R

In privacytoolsproject/PSI-Library: Differentially Private Statistical Releases for Privacy Preservation

library(PSIlence)
context('utilities-noise-generation')

test_that('Uniform distribution drawing functions as expected', {
    uniformRepeatable1 <- dpUnif(n=1, seed=75436)
    uniformRepeatable2 <- dpUnif(n=1, seed=75436)
    expect_equal(uniformRepeatable1, uniformRepeatable2)
    
    n <- 5
    uniformList <- dpUnif(n=n)
    expect_length(uniformList, n)
    expect_equal(sum(uniformList<1), n)
    expect_equal(sum(uniformList>0), n)
})

test_that('Laplace noise generation works when scale and number of noise parameters have varying dimensions',{
    # Case 1: generate a single noisy value from a Laplace distributions
    expect_equal(qLap(p=0.4, mu=0, b=3), 3*log(2 * 0.4))
    
    # Case 2: generate multiple noisy values, with vector of probabilities and single scaling parameter
    p <- c(0.4, 0.7)
    scale <- 3
    noise <- qLap(p, mu=0, b=scale)
    
    expect_equal(noise[1], scale*log(2*p[1]))
    expect_equal(noise[2], - scale*log(2-2 * p[2]))
    
    # Case 3: generate multiple noisy values, with vector of probabilities and vector of scaling parameters
    scale <- c(1, 5)
    noise <- qLap(p, mu=0, b=scale)
    
    expect_equal(noise[1], scale[1]*log(2*p[1]))
    expect_equal(noise[2], - scale[2]*log(2-2*p[2]))
})