tests/testthat/test-LDA.R

In sentopics: Tools for Joint Sentiment and Topic Analysis of Textual Data

context("testing LDA")
toks <- ECB_press_conferences_tokens[1:10]

LDA <- LDA(toks)

test_that("LDA works", {
  expect_output(
    print(LDA),
    "An LDA model with 5 topics. Currently fitted by 0 Gibbs sampling iterations."
  )
  LDA <- fit(LDA, 2, displayProgress = FALSE)
  expect_true(check_integrity(LDA, fast = FALSE))
  expect_s3_class(LDA, "LDA")

  LDA2 <- fit(LDA, 2, nChains = 2, displayProgress = FALSE)
  LDA2 <- fit(LDA2, 2, displayProgress = FALSE)
  expect_identical(attr(LDA2, "containedClass"), "LDA")

  expect_true(all(sapply(LDA2, check_integrity, fast = FALSE)))
})

LDA <- fit(LDA, 2, displayProgress = FALSE)

test_that("functions works", {
  expect_s3_class(top_words(LDA), "data.table")
  expect_true(is.matrix(top_words(LDA, output = "matrix")))
  expect_silent(p <- top_words(LDA, output = "plot"))
  expect_s3_class(p, "ggplot")
  expect_silent(print(p))
  expect_silent(m <- melt(LDA))
  expect_s3_class(m, "data.table")
  expect_silent(m <- data.table::melt(LDA))
  expect_s3_class(m, "data.table")
  # expect_equal(ncol(m), 6)
  skip_if_not_installed("plotly")
  expect_silent(p <- plot(LDA))
  expect_s3_class(p, "plotly")
  expect_silent(print(p))
})

test_that("test convergence", {
  vocab <- generateVocab(
    nTopics = 5,
    nSentiments = 1,
    nWords = 5,
    nCommonWords = 0,
    betaDirichlet = 10000000
  )
  toks <- generateDocuments(
    vocab,
    nDocs = 100,
    L1prior = 1,
    nWords = 100,
    nClass = 1
  )
  LDA <- LDA(toks)

  LDA <- fit(LDA, 100, nChains = 10, displayProgress = FALSE)
  expect_lte(mean(distToGenerative(LDA, vocab)), .15)
})

test_that("misc", {
  expect_length(dim(LDA$phi), 2)
  expect_length(dim(as.sentopicsmodel(LDA)$phi), 3)
  expect_length(dim(as.LDA(as.sentopicsmodel(LDA))$phi), 2)
})