tests/testthat/test_5_Tasks.R

In OscarKjell/text: Analyses of Text using Transformers Models from HuggingFace, Natural Language Processing and Machine Learning

library(testthat)
library(tibble)
library(text)

context("Testing tasks")


test_that("textClassify tests", {
  skip_on_cran()

  # String example help(textClassify) help(textClassify)
  print("textClassify")
  sen1 <- textClassifyPipe(
    x = "I like you. I love you",
    model = "distilbert-base-uncased-finetuned-sst-2-english",
    force_return_results = TRUE,
    return_all_scores = FALSE,
    function_to_apply = "none"
  )
  expect_equal(sen1$score_x, 4.67502, tolerance = 0.001)
  textModelsRemove("distilbert-base-uncased-finetuned-sst-2-english")
  #
})

test_that("textGeneration test", {
  skip_on_cran()

  print("textGeneration")
  generated_text <- textGeneration(
    x = "The meaning of life is",
    model = "gpt2",
    device = "cpu",
    tokenizer_parallelism = FALSE,
    logging_level = "warning",
    force_return_results = FALSE,
    return_tensors = FALSE,
    return_full_text = TRUE,
    clean_up_tokenization_spaces = FALSE,
    prefix = "",
    handle_long_generation = "hole",
    set_seed = 22L
  )
  # torch 1.11
  # expect_equal(generated_text$x_generated, "The meaning of life is to live for
  # the sake of one's children, and to take the name or life of another, and not as
  # the child of a dead person. It is a name, a life, and not as the son, daughter")
  expect_that(generated_text$x_generated, is_a("character"))

  # Use `max_new_tokens` and convert set_seed to integer
  print("textGeneration_2")
  generated_text2 <- text::textGeneration(
    x = "The meaning of life is",
    model = "gpt2",
    device = "cpu",
    tokenizer_parallelism = FALSE,
    max_new_tokens = 2,
    logging_level = "warning",
    force_return_results = FALSE,
    return_tensors = FALSE,
    return_full_text = TRUE,
    clean_up_tokenization_spaces = FALSE,
    prefix = "",
    handle_long_generation = "hole",
    set_seed = 22
  )
  expect_that(generated_text2$x_generated, is_a("character"))

  # Use `max_length`
  print("textGeneration_3")
  generated_text3 <- text::textGeneration(
    x = "The meaning of life is",
    model = "gpt2",
    device = "cpu",
    tokenizer_parallelism = FALSE,
    max_length = 20,
    logging_level = "warning",
    force_return_results = FALSE,
    return_tensors = FALSE,
    return_full_text = TRUE,
    clean_up_tokenization_spaces = FALSE,
    prefix = "",
    handle_long_generation = "hole",
    set_seed = 22L
  )
  expect_that(generated_text3$x_generated, is_a("character"))

  # Use to small a `max_length` without setting `max_new_tokens`
  print("textGeneration_4")
  text::textGeneration(
    x = "The meaning of life is",
    model = "gpt2",
    device = "cpu",
    tokenizer_parallelism = FALSE,
    max_length = 3L,
    max_new_tokens = NULL,
    logging_level = "warning",
    force_return_results = FALSE,
    return_tensors = FALSE,
    return_full_text = TRUE,
    clean_up_tokenization_spaces = FALSE,
    prefix = "",
    handle_long_generation = "hole",
    set_seed = 22L
  ) %>% expect_error(regexp = 'ValueError\\: Input length')

  # Use `min_length`
  print("textGeneration_5")
  generated_text5 <- text::textGeneration(
    x = "The meaning of life is",
    model = "gpt2",
    device = "cpu",
    tokenizer_parallelism = FALSE,
    max_length = NULL,
    min_length = 20L,
    max_new_tokens = NULL,
    logging_level = "warning",
    force_return_results = FALSE,
    return_tensors = FALSE,
    return_full_text = TRUE,
    clean_up_tokenization_spaces = FALSE,
    prefix = "",
    handle_long_generation = "hole",
    set_seed = 22L
  )
  expect_that(generated_text5$x_generated, is_a("character"))

  # Use `min_new_tokens`
  print("textGeneration_6")
  generated_text6 <- text::textGeneration(
    x = "The meaning of life is",
    model = "gpt2",
    device = "cpu",
    tokenizer_parallelism = FALSE,
    min_new_tokens = 10,
    max_new_tokens = NULL,
    logging_level = "warning",
    force_return_results = FALSE,
    return_tensors = FALSE,
    return_full_text = TRUE,
    clean_up_tokenization_spaces = FALSE,
    prefix = "",
    handle_long_generation = "hole",
    set_seed = 22L
  )
  expect_that(generated_text6$x_generated, is_a("character"))

  # All length parameters set to NULL
  print("textGeneration_7")
  generated_text7 <- text::textGeneration(
    x = "The meaning of life is",
    model = "gpt2",
    device = "cpu",
    tokenizer_parallelism = FALSE,
    max_length = NULL,
    min_length = NULL,
    max_new_tokens = NULL,
    min_new_tokens = NULL,
    logging_level = "warning",
    force_return_results = FALSE,
    return_tensors = FALSE,
    return_full_text = TRUE,
    clean_up_tokenization_spaces = FALSE,
    prefix = "",
    handle_long_generation = "hole",
    set_seed = 22L
  )
  expect_that(generated_text7$x_generated, is_a("character"))

  # Return token IDs
  print("textGeneration_8")
  generated_text8 <- text::textGeneration(
    x = "The meaning of life is",
    model = "gpt2",
    device = "cpu",
    tokenizer_parallelism = FALSE,
    logging_level = "warning",
    force_return_results = FALSE,
    return_tensors = TRUE,
    return_full_text = FALSE,
    clean_up_tokenization_spaces = FALSE,
    prefix = "",
    handle_long_generation = "hole",
    set_seed = 22L
  )
  textModelsRemove("gpt2")
  expect_equal(generated_text8$generated_token_ids[1], 464)
  expect_that(generated_text8$generated_token_ids[1], is_a("integer"))
})


test_that("textNER test", {
  skip_on_cran()

  print("textNER")
  ner_example <- textNER("Arnes plays football with Daniel",
    mode = "dslim/bert-base-NER",
    force_return_results = FALSE
  )
  ner_example

  expect_equal(ner_example$x_NER$score[1], 0.9987748, tolerance = 0.001)

  print("textNER_2")
  ner_example2 <- textNER(Language_based_assessment_data_8[1:2, 1],
    mode = "dslim/bert-base-NER",
    force_return_results = FALSE
  )
  ner_example2
  expect_equal(ner_example2$satisfactiontexts_NER$score[2], 0.976, tolerance = 0.01)
  textModelsRemove("dslim/bert-base-NER")
})

test_that("textSum test", {
  skip_on_cran()

  print("textSum")
  sum_examples <- textSum(Language_based_assessment_data_8[1:2, 1:2],
    min_length = 2L,
    max_length = 4L
  )
  sum_examples
  expect_that(sum_examples$sum_satisfactiontexts, is_a("character"))
  expect_equal(sum_examples$sum_satisfactiontexts[1], "I am not")
})


test_that("textQA test", {
  skip_on_cran()

  print("textQA")
  qa_examples <- textQA(
    question = "Which colour have trees?",
    context = "Trees are mostly green and like water"
  )
  expect_equal(qa_examples$score, .9384409, tolerance = 0.0001)
  expect_equal(qa_examples$start, 17)
  expect_equal(qa_examples$end, 22)
  expect_equal(qa_examples$answer, "green")
})



test_that("textZeroShot test", {
  skip_on_cran()

  print("textZeroShot")
  ZeroShot_example <- text::textZeroShot(
    sequences = c("I play football", "The forrest is wonderful"),
    candidate_labels = c("sport", "nature", "research"),
    # model = "facebook/bart-large-mnli"
    model = "okho0653/distilbert-base-uncased-zero-shot-sentiment-model"
  )

  testthat::expect_equal(ZeroShot_example$scores_x_1[1], 0.3341856, tolerance = 0.00001)
  textModelsRemove("okho0653/distilbert-base-uncased-zero-shot-sentiment-model")
})

test_that("textTranslate test", {
  skip_on_cran()

  print("textTranslate")
  textModels()
  translation_example <- text::textTranslate(
    Language_based_assessment_data_8[1, 1:2],
    source_lang = "en",
    target_lang = "fr",
    model = "t5-small",
    max_length = 400
  )

  testthat::expect_that(translation_example$en_to_fr_satisfactiontexts, testthat::is_a("character"))
  testthat::expect_equal(
    translation_example$en_to_fr_satisfactiontexts[1],
    "Je ne suis pas satisfait de ma vie, je suis reconnaissante de ce que j'ai et de ce que je suis, car la situation peut toujours être pire. Je veux une carrière et un diplôme, je veux perdre de poids et je n'ai pas encore atteint ces objectifs."
  )
  textModelsRemove("t5-small")
})