tests/testthat/test-foehnix_functions.R

In retostauffer/Rfoehnix: Objective Foehn Diagnosis

library("testthat")
library("foehnix")


test_that("Convert windsector information", {

    # Allowed: NULL, matrix, list, data.frame
    expect_is(foehnix:::windsector_convert(NULL), "NULL")
    expect_is(foehnix:::windsector_convert(matrix(c(10, 30, 90, 140), byrow = TRUE, ncol = 2)), "list")
    expect_is(foehnix:::windsector_convert(matrix(c(10, 30, 90, 140), byrow = TRUE, ncol = 2,
                                        dimnames = list(c("A", "B"), c("from", "to")))), "list")
    expect_is(foehnix:::windsector_convert(list(c(10, 30), c(90, 140))), "list")
    expect_is(foehnix:::windsector_convert(list(A = c(10, 30), B = c(90, 140))), "list")
    expect_is(foehnix:::windsector_convert(data.frame(from = c(30, 90), to = c(30, 140))), "list")
    expect_is(foehnix:::windsector_convert(structure(data.frame(from = c(30, 90), to = c(30, 140)),
                                           row.names = c("foo", "bar"))), "list")

    # Not allowed
    expect_error(foehnix:::windsector_convert(c(0, 100)))
    expect_error(foehnix:::windsector_convert(100, 200))
})


test_that("Foehnix functions: standardize/destandardize", {

    # Wrong inputs
    expect_error(standardize(3))
    expect_error(standardize(data.frame(a = c(1,2), b = c(2,3))))
    expect_error(standardize(3, "foo"))

    # Take the airquality data set to test standardization
    expect_silent(X <- as.matrix(na.omit(airquality)))
    expect_silent(S <- standardize(X))
    expect_is(S, c("standardized", "matrix"))
    expect_identical(dim(X), dim(S))
    expect_is(attr(S, "scaled:center"), "numeric")
    expect_is(attr(S, "scaled:scale"), "numeric")
    expect_identical(attr(S, "scaled:center"), apply(X, 2, mean))
    expect_identical(attr(S, "scaled:scale"), apply(X, 2, sd))
    expect_equal(apply(S, 2, mean), structure(rep(0, ncol(S)), names = colnames(S)))
    expect_equal(apply(S, 2, sd),   structure(rep(1, ncol(S)), names = colnames(S)))

    # Extract attributes
    expect_equal(center(S), apply(X, 2, mean))
    expect_equal(scale(S),  apply(X, 2, sd))

    # Destandardize
    expect_silent(D <- destandardize(S))
    expect_identical(dim(D), dim(X))
    expect_equal(unclass(D), unclass(X))

    # Checking standardization
    expect_false(is.standardized(X))
    expect_true(is.standardized(S))
    expect_false(is.standardized(D))


})

test_that("Foehnix functions: destandardize coefficients", {

    # Generate a fake standardized matrix
    set.seed(10)
    S <- matrix(c(rnorm(500, 10, 2), rnorm(500, 5, 3), rnorm(500, -5, 1)),
                ncol = 3, dimnames = list(NULL, c("(Intercept)", "A", "B")))
    S <- standardize(S)
    # Fake scaled:center and scaled:scale
    attr(S, "scaled:center") <- structure(c(10, 5, -20), names = colnames(S))
    attr(S, "scaled:scale")  <- structure(c(1, 2, 5), names = colnames(S))
    expect_true(is.standardized(S))

    # Check if conversion works: both standardized with an
    # offset of 20 and a scale of 2. The mean is meaningless
    # for the coefficients except for '(Intercept)'. 
    beta <- structure(matrix(c(2, -5, -5), ncol = 1), dimnames = list(c("(Intercept)", "A", "B"), NULL))
    expect_silent(beta2 <- destandardize_coefficients(beta, S))
    # Manually de-scaling. 
    beta3 <- vector("numeric", length(beta))
    beta3[2:3] <- as.vector(beta[2:3]) / attr(S, "scaled:scale")[2:3]
    beta3[1]   <- beta[1] - sum(beta[2:3] / attr(S, "scaled:scale")[2:3] * attr(S, "scaled:center")[2:3])
    # Compare
    expect_equal(as.vector(beta2), as.vector(beta3))

})