def test_question_answering_visualize():
qa_explainer = QuestionAnsweringExplainer(DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER)
explainer_question = "what is his name ?"
explainer_text = "his name is Bob"
qa_explainer(explainer_question, explainer_text)
qa_explainer.visualize()
def test_question_answering_encode():
qa_explainer = QuestionAnsweringExplainer(DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER)
_input = "this is a sample of text to be encoded"
tokens = qa_explainer.encode(_input)
assert isinstance(tokens, list)
assert tokens[0] != qa_explainer.cls_token_id
assert tokens[-1] != qa_explainer.sep_token_id
assert len(tokens) >= len(_input.split(" "))
def test_question_answering_visualize_save_append_html_file_ending():
qa_explainer = QuestionAnsweringExplainer(DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER)
explainer_question = "what is his name ?"
explainer_text = "his name is Bob"
qa_explainer(explainer_question, explainer_text)
html_filename = "./test/qa_test"
qa_explainer.visualize(html_filename)
assert os.path.exists(html_filename + ".html")
os.remove(html_filename + ".html")
def test_question_answering_decode():
qa_explainer = QuestionAnsweringExplainer(DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER)
explainer_question = "what is his name ?"
explainer_text = "his name is bob"
input_ids, _, _ = qa_explainer._make_input_reference_pair(
explainer_question, explainer_text)
decoded = qa_explainer.decode(input_ids)
assert decoded[0] == qa_explainer.tokenizer.cls_token
assert decoded[-1] == qa_explainer.tokenizer.sep_token
assert (" ".join(decoded[1:-1]) == explainer_question.lower() + " [SEP] " +
explainer_text.lower())
def test_question_answering_explainer_init_attribution_type_error():
with pytest.raises(AttributionTypeNotSupportedError):
QuestionAnsweringExplainer(
DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER,
attribution_type="UNSUPPORTED",
)
def test_question_answering_word_attributions_input_ids_not_calculated():
qa_explainer = QuestionAnsweringExplainer(
DISTILBERT_QA_MODEL, DISTILBERT_QA_TOKENIZER
)
with pytest.raises(ValueError):
qa_explainer.word_attributions
def test_question_answering_explainer_init_distilbert():
qa_explainer = QuestionAnsweringExplainer(
DISTILBERT_QA_MODEL, DISTILBERT_QA_TOKENIZER
)
assert qa_explainer.attribution_type == "lig"
assert qa_explainer.attributions is None
assert qa_explainer.position == 0
def test_question_answering_end_pos():
qa_explainer = QuestionAnsweringExplainer(DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER)
explainer_question = "what is his name ?"
explainer_text = "his name is Bob"
qa_explainer(explainer_question, explainer_text)
end_pos = qa_explainer.end_pos
assert end_pos == 10
def test_question_answering_predicted_answer():
qa_explainer = QuestionAnsweringExplainer(DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER)
explainer_question = "what is his name ?"
explainer_text = "his name is Bob"
qa_explainer(explainer_question, explainer_text)
predicted_answer = qa_explainer.predicted_answer
assert predicted_answer == "bob"
def test_question_answering_word_attributions():
qa_explainer = QuestionAnsweringExplainer(DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER)
explainer_question = "what is his name ?"
explainer_text = "his name is bob"
word_attributions = qa_explainer(explainer_question, explainer_text)
assert isinstance(word_attributions, dict)
assert "start" in word_attributions.keys()
assert "end" in word_attributions.keys()
assert len(word_attributions["start"]) == len(word_attributions["end"])
def test_question_answering_start_pos_input_ids_not_calculated():
qa_explainer = QuestionAnsweringExplainer(DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER)
with pytest.raises(InputIdsNotCalculatedError):
qa_explainer.start_pos
def xtest_question_answering_custom_internal_batch_size():
qa_explainer = QuestionAnsweringExplainer(DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER)
explainer_question = "what is his name ?"
explainer_text = "his name is Bob"
qa_explainer(explainer_question, explainer_text, internal_batch_size=1)
def xtest_question_answering_custom_steps():
qa_explainer = QuestionAnsweringExplainer(DISTILBERT_QA_MODEL,
DISTILBERT_QA_TOKENIZER)
explainer_question = "what is his name ?"
explainer_text = "his name is Bob"
qa_explainer(explainer_question, explainer_text, n_steps=1)
