tests/testthat/test.estimation.R

In modal-inria/cfda: Categorical Functional Data Analysis

# Author: Quentin Grimonprez

context("Markov estimation")

set.seed(42)

test_that("completeStatetable does not change a square statetable", {
  aux <- matrix(c(0, 2, 3, 4, 1, 0, 3, 4, 1, 2, 0, 4, 1, 2, 3, 0), nrow = 4, dimnames = list(1:4, 1:4))

  out <- completeStatetable(aux)

  expect_equal(out, aux)
})


test_that("completeStatetable completes one missing row", {
  # middle
  aux <- matrix(c(0, 3, 4, 1, 3, 4, 1, 0, 4, 1, 3, 0), nrow = 3, dimnames = list(c(1, 3, 4), 1:4))

  out <- completeStatetable(aux)

  expectedOut <- rbind(aux[1, ], rep(0, 4), aux[2:3, ])
  rownames(expectedOut) <- 1:4

  expect_equal(out, expectedOut)


  # first
  aux <- matrix(c(2, 3, 4, 0, 3, 4, 2, 0, 4, 2, 3, 0), nrow = 3, dimnames = list(c(2, 3, 4), 1:4))

  out <- completeStatetable(aux)

  expectedOut <- rbind(rep(0, 4), aux)
  rownames(expectedOut) <- 1:4

  expect_equal(out, expectedOut)


  # last
  aux <- matrix(c(0, 2, 3, 1, 0, 3, 1, 2, 0, 1, 2, 3), nrow = 3, dimnames = list(c(1, 2, 3), 1:4))

  out <- completeStatetable(aux)

  expectedOut <- rbind(aux, rep(0, 4))
  rownames(expectedOut) <- 1:4

  expect_equal(out, expectedOut)
})


test_that("completeStatetable completes several missing rows", {
  aux <- matrix(c(0, 3, 1, 3, 1, 0, 1, 3), nrow = 2, dimnames = list(c(1, 3), 1:4))

  out <- completeStatetable(aux)

  expectedOut <- rbind(aux[1, ], rep(0, 4), aux[2, ], rep(0, 4))
  rownames(expectedOut) <- 1:4

  expect_equal(out, expectedOut)
})

test_that("completeStatetable completes several missing rows (2 lasts)", {
  aux <- matrix(c(0, 3, 1, 3, 1, 0, 1, 3), nrow = 2, dimnames = list(c(1, 2), 1:4))

  out <- completeStatetable(aux)

  expectedOut <- rbind(aux[1, ], aux[2, ], rep(0, 4), rep(0, 4))
  rownames(expectedOut) <- 1:4

  expect_equal(out, expectedOut)
})

test_that("completeStatetable completes several missing rows non integer labels", {
  aux <- matrix(c(0, 3, 1, 3, 1, 0, 1, 3), nrow = 2, dimnames = list(c("A", "B"), c("A", "B", "C", "D")))

  out <- completeStatetable(aux)

  expectedOut <- rbind(aux[1, ], aux[2, ], rep(0, 4), rep(0, 4))
  rownames(expectedOut) <- c("A", "B", "C", "D")

  expect_equal(out, expectedOut)
})

test_that("estimateT estimates well", {
  data <- data.frame(
    id = rep(c(1, 2), each = 5),
    time = c(1, 3, 7, 9, 10, 1, 2, 5, 6, 10),
    state = c(1, 2, 1, 3, 1, 1, 3, 2, 1, 3)
  )

  out <- estimateT(data)

  expectedOut <- c(2.25, 2.5, 2)

  expect_equivalent(out, expectedOut)
})


test_that("estimate_Markov estimates well", {
  K <- 4
  PJK <- matrix(1 / 3, nrow = K, ncol = K) - diag(rep(1 / 3, K))
  lambda_PJK <- c(1, 1, 1, 1)
  d_JK <- generate_Markov(n = 500, K = K, P = PJK, lambda = lambda_PJK, Tmax = 30)

  mark <- estimate_Markov(d_JK)

  expect_lte(sqrt(mean((mark$lambda - lambda_PJK)^2)), 0.06)
  expect_lte(sqrt(mean((mark$P - PJK)^2)), 0.02)
})

test_that("estimate_Markov estimates well", {
  K <- 4
  PJK <- matrix(1 / 3, nrow = K, ncol = K) - diag(rep(1 / 3, K))
  lambda_PJK <- c(1, 1, 1, 1)
  d_JK <- generate_Markov(n = 500, K = K, P = PJK, lambda = lambda_PJK, Tmax = 30)

  mark <- estimate_Markov(d_JK)

  expect_lte(sqrt(mean((mark$lambda - lambda_PJK)^2)), 0.06)
  expect_lte(sqrt(mean((mark$P - PJK)^2)), 0.02)
})

test_that("estimate_Markov works with missing transitions", {
  K <- 4
  d_JK <- data.frame(id = rep(1:10, each = 2),
                     time = rep(0:1, 10),
                     state = rep(c("C", "D"), 10))
  d_JK$state[2] <- "T"
  d_JK$state[3:4] <- c("D", "T")
  d_JK$state[6] <- "C"

  mark <- estimate_Markov(d_JK)
  lam <- c(1, 1, NaN)
  names(lam) <- c("C", "D", "T")

  p <- matrix(c(0, 0, NaN, 0.875, 0, NaN, 0.125, 1, NaN), nrow = , ncol = 3,
              dimnames = list(c("C", "D", "T"), c("C", "D", "T")))
  expect_equal(mark$lambda, lam)
  expect_equal(mark$P, p)
})

test_that("plot_Markov does not produce warnings", {
  K <- 4
  PJK <- matrix(1 / 3, nrow = K, ncol = K, dimnames = list(1:4, 1:4)) - diag(rep(1 / 3, K))
  lambda_PJK <- c(1, 1, 1, 1)

  dat <- list(P = PJK, lambda = lambda_PJK)
  class(dat) <- "Markov"

  expect_warning(plot(dat), regexp = NA)
})