tests/testthat/test-machine_learning.R

In THargreaves/online-oceanarium: Online Oceanarium

test_that("LinearRegression provides correct results", {
    methods <- c(
        "sherman-morrison",
        "woodbury"
    )
    for (method in methods) {
        X <- matrix(c(3, 1, 4, 1, 5, 9, 2, 6), ncol = 2)
        y <- c(2, 7, 1, 8)
        # Initialisation
        reg <- LinearRegression$new(X, y, method = method)
        expect_equal(reg$value, unname(coef(lm(y ~ X))))
        # Updating
        X_new <- matrix(c(5, 3, 5, 8), ncol = 2)
        y_new <- c(2, 8)
        reg$update(X_new, y_new)
        expect_equal(
            reg$value,
            unname(coef(lm(c(y, y_new) ~ rbind(X, X_new))))
        )
        # No intercept
        reg <- LinearRegression$new(X, y, fit_intercept = FALSE,
                                    method = method)
        expect_equal(reg$value, unname(coef(lm(y ~ X + 0))))
    }
})

test_that("LinearRegression handles errors", {
    # Invalid method
    X <- matrix(c(3, 1, 4, 1, 5, 9, 2, 6), ncol = 2)
    y <- c(2, 7, 1, 8)
    expect_error(LinearRegression$new(X, y, method = "spam"))
    # Invalid data
    expect_error(LinearRegression$new(NULL, y))
    expect_error(LinearRegression$new(X, NULL))
    expect_error(LinearRegression$new(X, y[-1]))
    # Invalid update
    reg <- LinearRegression$new(X, y)
    X_new <- matrix(c(5, 3, 5), ncol = 3)
    expect_error(reg$update(X_new))
})