tests/testthat/test_geneig.R

In multivarious: Extensible Data Structures for Multivariate Analysis

library(testthat)
library(Matrix) # For diagonal matrix operations and checks

# Define a known symmetric matrix A and a positive definite matrix B
A <- matrix(c(4, 1, 1, 2), nrow=2, byrow=TRUE)
B <- matrix(c(6, 2, 2, 5), nrow=2, byrow=TRUE)  # B must be symmetric and positive definite


test_that("geigen method returns correct results", {
  result <- geneig(A = A, B = B, ncomp=2, method="geigen")
  expect_equal(dim(result$vectors), c(2, 2))
  expect_equal(length(result$values), 2)
  # Eigenvalues can be complex for geigen, check real part if needed or just existence
  # expect_true(all(Re(result$values) > 0)) # This might not hold for general A
})

test_that("robust method equivalent test using geigen", {
  # Changed method from "robust" to "geigen" as "robust" is removed
  result <- geneig(A = A, B = B, ncomp=2, method="geigen")
  expect_equal(dim(result$vectors), c(2, 2))
  expect_equal(length(result$values), 2)
  # Cannot guarantee positive values for geigen
  # expect_true(all(result$values > 0))
})

test_that("sdiag method equivalent test using geigen", {
  # Changed method from "sdiag" to "geigen" as "sdiag" is removed
  result <- geneig(A = A, B = B, ncomp=2, method="geigen")
  expect_equal(dim(result$vectors), c(2, 2))
  expect_equal(length(result$values), 2)
  # Cannot guarantee positive values for geigen
  # expect_true(all(result$values > 0))
})

test_that("primme method returns correct results", {
  skip_if_not_installed("PRIMME")  # Skip if PRIMME is not available
  # Suppress expected warning about switching to dense backend for full decomposition
  result <- suppressWarnings(geneig(A = A, B = B, ncomp=2, method="primme", which="LA"))
  expect_equal(dim(result$vectors), c(2, 2))
  expect_equal(length(result$values), 2)
  expect_true(all(result$values > 0))
})

test_that("non-square matrices are handled", {
  non_square_A <- matrix(1:6, nrow=2)
  non_square_B <- matrix(1:6, nrow=2)
  
  expect_error(geneig(A = non_square_A, B = non_square_B, ncomp=2, method="geigen"))
})

test_that("negative and very small eigenvalues in B are handled in geigen", {
  B_with_negative <- matrix(c(4, 1, 1, -2), nrow=2, byrow=TRUE)
  # Changed method from "sdiag" to "geigen"
  # Suppress expected warning about negative eigenvalues
  result <- suppressWarnings(geneig(A = A, B = B_with_negative, ncomp=2, method="geigen"))
  expect_equal(dim(result$vectors), c(2, 2))
  # Cannot guarantee positive values for geigen
  # expect_true(all(result$values > 0))
})