def test_duplicate_ids_not_allowed(self, mock_download_dataset):
self.write_questions(HotpotQAType.TRAIN,
[TRAIN_QUESTION, TRAIN_QUESTION])
with self.assertRaises(AssertionError):
dm.HOTPOT_QA(HotpotQAType.TRAIN)
mock_download_dataset.assert_called_once_with(Collection.HOTPOT_QA,
ANY) # noqa: E501
def test_load_in_test_fullwiki_format(self, mock_download_dataset):
self.write_questions(HotpotQAType.TEST_FULLWIKI,
[TEST_FULLWIKI_QUESTION]) # noqa: E501
df = dm.HOTPOT_QA(HotpotQAType.TEST_FULLWIKI)
expected_df = pd.DataFrame(
json.loads("""[
{
"id": "5ab5072e5542990594ba9cda",
"question": "Test question?",
"answer": null,
"gold_paragraphs": [],
"supporting_facts": [],
"context": [
[
"The Rolling Stone Album Guide",
["Sent 1.", " Sent 2."]
],
[
"Fear and Loathing at Rolling Stone",
["Sent 3."]
]
],
"question_type": null,
"question_level": null
}
]"""))
pd.testing.assert_frame_equal(df, expected_df)
mock_download_dataset.assert_called_once_with(Collection.HOTPOT_QA,
ANY) # noqa: E501
def test_load_multiple_questions(self, mock_download_dataset):
similar_question = deepcopy(TRAIN_QUESTION)
similar_question["_id"] = "aaaabbbbccccdddd!2"
self.write_questions(HotpotQAType.TRAIN,
[TRAIN_QUESTION, similar_question])
df = dm.HOTPOT_QA(HotpotQAType.TRAIN)
mock_download_dataset.assert_called_once_with(Collection.HOTPOT_QA,
ANY) # noqa: E501
self.assertEqual(len(df), 2)
def main():
df = dm.HOTPOT_QA(HotpotQAType.DEV_DISTRACTOR)
print(df)
print("\n")
df = df.sample(n=1)
row = next(df.iterrows())[1]
gold_paragraphs = row.gold_paragraphs
print("Question: ", row.question)
print("Answer: ", row.answer, "\n")
for i, paragraph in enumerate(gold_paragraphs):
print("Paragraph {}) {}\n".format(chr(ord('A') + i), paragraph))
def test_load_in_dev_fullwiki_format(self, mock_download_dataset):
self.write_questions(HotpotQAType.DEV_FULLWIKI,
[DEV_FULLWIKI_QUESTION]) # noqa: E501
df = dm.HOTPOT_QA(HotpotQAType.DEV_FULLWIKI)
expected_df = pd.DataFrame(
json.loads("""[
{
"id": "5a899013554299515336131a",
"question": "Some question",
"answer": "Dallas",
"gold_paragraphs": [
"Sent good 3."
],
"supporting_facts": [
["Limitless (EP)", 0],
["Crown the Empire", 0]
],
"context": [
[
"The Resistance: Rise of The Runaways",
["Sent 1", " Sent 2.", " Sent 3."]
],
[
"Reign of Terror (Capture the Crown album)",
["Sent 4."]
],
[
"Retrograde (Crown the Empire album)",
["Sent 5.", " Sent 6."]
],
[
"Roots (Sepultura album)",
["Sent 7.", " Sent 8.", " Sent 9."]
],
[
"Forest Stream",
["Sent 10.", " Sent 11.", " Sent 12.", " Sent 13.",
" Sent 14.", " Sent 15."]
],
[
"The Crown (band)",
["Sent 16."]
],
[
"The Fallout (Crown the Empire album)",
["Sent 17.", " Sent 18."]
],
[
"Crown the Empire discography",
[" Sent 19."]
],
[
"Crown the Empire - is missing",
["Sent good 1.", " Sent good 2."]
],
[
"Limitless (EP)",
["Sent good 3."]
]
],
"question_type": "bridge",
"question_level": "medium"
}
]"""))
pd.testing.assert_frame_equal(df, expected_df)
mock_download_dataset.assert_called_once_with(Collection.HOTPOT_QA,
ANY) # noqa: E501
def test_empty_dataset(self, mock_download_dataset):
self.write_questions(HotpotQAType.TRAIN, [])
df = dm.HOTPOT_QA(HotpotQAType.TRAIN)
mock_download_dataset.assert_called_once_with(Collection.HOTPOT_QA,
ANY) # noqa: E501
self.assertEqual(len(df), 0)