tests/testthat/test-utils.R

In text2vec: Modern Text Mining Framework for R

context("text2vec utils - tokenization, etc.")

txt = tolower(movie_review$review[[1]])
N = 20
it = itoken(movie_review$review[1:N], ids = movie_review$id[1:N], progressbar = FALSE)
dtm = create_dtm(it, hash_vectorizer(2**8), "TsparseMatrix")

txt_first_10 = c("with", "all", "this", "stuff",
             "going", "down", "at", "the",
             "moment", "with")

test_that("word_tokenizer ", {
  tokens = word_tokenizer(txt)[[1]]
  expect_equal(length(tokens), 438)
  expect_equal(tokens[1:10], txt_first_10)
  # non ASCII symbols
  tokens = word_tokenizer("one, two. Three! four")[[1]]
  expect_equal(tokens, c("one", "two", "Three", "four"))
})

test_that("space_tokenizer ", {
  tokens = space_tokenizer(txt)[[1]]
  expect_equal(length(tokens), 433)
  expect_equal(tokens[1:10], txt_first_10)
  expect_error(space_tokenizer("one, two. Three! four", "[[:punct:]]|\\s+")[[1]])
})

test_that("char_tokenizer ", {
  txt = "aaabbc!."
  tokens = char_tokenizer(txt)[[1]]
  expect_equal(length(tokens), nchar(txt))
})

test_that("split_into ", {
  n_splits = 3
  splits = split_into(1:N, n_splits)
  expect_equal(length(splits), n_splits)
  max_chunk_len = max(vapply(splits, length, 0L))
  expect_lte(max_chunk_len, ceiling(N / n_splits))
})

test_that("rbind_dgTMatrix ", {
  it1 = itoken(movie_review$review[1:10], ids = movie_review$id[1:10], progressbar = FALSE)
  dtm1 = create_dtm(it1, hash_vectorizer(2**8), "TsparseMatrix")
  it2 = itoken(movie_review$review[11:20], ids = movie_review$id[11:20], progressbar = FALSE)
  dtm2 = create_dtm(it2, hash_vectorizer(2**8), "TsparseMatrix")
  expect_equal(Matrix::uniqTsparse(text2vec:::rbind_dgTMatrix(dtm1, dtm2)),
               Matrix::uniqTsparse(dtm))
})
test_that("as.lda_c ", {
  K = 100
  train_tokens = tolower(movie_review$review[1:K])
  train_tokens =  word_tokenizer(train_tokens)

  it_train = itoken(train_tokens,
                    ids = movie_review$id[1:K])

  vocab = create_vocabulary(it_train)
  vocab = prune_vocabulary(vocab, term_count_min = 10, doc_proportion_max = 0.1)
  dtm = create_dtm(it_train, vocab_vectorizer(vocab))
  rs = Matrix::rowSums(sign(dtm))
  expect_equal(sum(rs == 0) , 2)
  dtm_lda_c = as.lda_c(dtm)
  expect_equal(names(dtm_lda_c), movie_review$id[1:K])
  expect_equal(rs, vapply(dtm_lda_c, ncol, 0L))
  expect_equal(Matrix::rowSums(dtm), vapply(dtm_lda_c, function(x) sum(x[2, ]), 0L))
})