tests/testthat/test-pca.R

In cuda.ml: R Interface for the RAPIDS cuML Suite of Libraries

context("Principal Component Analysis")

cuda_ml_pca_model <- cuda_ml_pca(
  iris[, which(names(iris) != "Species")],
  n_components = 3
)

sklearn_pca_model <- sklearn$decomposition$PCA(n_components = 3L, copy = TRUE)
sklearn_pca_model$fit(sklearn_iris_dataset$data)

test_that("cuda_ml_pca() works as expected", {
  expect_equal(
    cuda_ml_pca_model$components, as.matrix(sklearn_pca_model$components_),
    tolerance = 1e-8, scale = 1
  )
  expect_equal(
    cuda_ml_pca_model$explained_variance,
    as.numeric(sklearn_pca_model$explained_variance_),
    tolerance = 1e-8, scale = 1
  )
  expect_equal(
    cuda_ml_pca_model$explained_variance_ratio,
    as.numeric(sklearn_pca_model$explained_variance_ratio_),
    tolerance = 1e-8, scale = 1
  )
  expect_equal(
    cuda_ml_pca_model$singular_values,
    as.numeric(sklearn_pca_model$singular_values_),
    tolerance = 1e-8, scale = 1
  )
  expect_equal(
    cuda_ml_pca_model$mean, as.numeric(sklearn_pca_model$mean_),
    tolerance = 1e-8, scale = 1
  )
})

test_that("cuda_ml_inverse_transform() works as expected for PCA models", {
  expect_equal(
    cuda_ml_pca_model$transformed_data %>%
      sklearn_pca_model$inverse_transform() %>%
      as.matrix(),
    cuda_ml_pca_model %>%
      cuda_ml_inverse_transform(cuda_ml_pca_model$transformed_data),
    tolerance = 1e-8, scale = 1
  )
})