def test_collect_matches2doc_ranker_driver_min_ranker():
doc = create_document_to_score_same_depth_level()
driver = SimpleCollectMatchesRankDriver()
executor = MinRanker()
driver.attach(executor=executor, pea=None)
import sys
min_value_30 = sys.maxsize
min_value_20 = sys.maxsize
for match in doc.matches:
if match.parent_id == '30':
if match.score.value < min_value_30:
min_value_30 = match.score.value
if match.parent_id == '20':
if match.score.value < min_value_20:
min_value_20 = match.score.value
assert min_value_30 < min_value_20
driver._traverse_apply([
doc,
])
assert len(doc.matches) == 2
assert doc.matches[0].id == '30'
assert doc.matches[0].score.value == pytest.approx(
(1. / (1. + min_value_30)), 0.0000001)
assert doc.matches[1].id == '20'
assert doc.matches[1].score.value == pytest.approx(
(1. / (1. + min_value_20)), 0.0000001)
for match in doc.matches:
# match score is computed w.r.t to doc.id
assert match.score.ref_id == doc.id
def test_chunk2doc_ranker_driver_traverse_apply_larger_range():
docs = [
create_chunk_chunk_matches_to_score(),
]
driver = SimpleChunk2DocRankDriver(traversal_paths=('cc', 'c'))
executor = MinRanker()
driver.attach(executor=executor, pea=None)
driver._traverse_apply(docs)
for doc in docs:
assert len(doc.matches) == 1
assert len(doc.chunks) == 1
chunk = doc.chunks[0]
assert len(chunk.matches) == 2
min_granularity_2 = chunk.matches[0].score.value
for idx, m in enumerate(chunk.matches):
# the score should be 1 / (1 + id * 2)
if m.score.value < min_granularity_2:
min_granularity_2 = m.score.value
assert m.score.value == pytest.approx(1. / (1 + float(m.id) * 2.),
0.0001)
assert m.score.ref_id == 101
match = doc.matches[0]
assert match.score.ref_id == 100
assert match.score.value == pytest.approx(1. / (1 + min_granularity_2),
0.0001)
def test_chunk2doc_ranker_driver_traverse_apply():
docs = [create_chunk_matches_to_score(), ]
driver = SimpleChunk2DocRankDriver(recur_range=(0, 1))
executor = MinRanker()
driver.attach(executor=executor, pea=None)
driver._traverse_apply(docs)
for doc in docs:
assert len(doc.matches) == 2
for idx, m in enumerate(doc.matches):
# the score should be 1 / (1 + id * 2)
assert m.score.value == pytest.approx(1. / (1 + m.id * 2.), 0.0001)
def test_chunk2doc_ranker_driver_traverse_apply(self):
docs = [
create_chunk_matches_to_score(),
]
driver = SimpleChunk2DocRankDriver(depth_range=(0, 1))
executor = MinRanker()
driver.attach(executor=executor, pea=None)
driver._traverse_apply(docs)
for doc in docs:
assert len(doc.matches) == 2
for idx, m in enumerate(doc.matches):
# the score should be 1 / (1 + id * 2)
self.assertAlmostEqual(m.score.value, 1. / (1 + m.id * 2.))
assert m.level_depth == 0
def test_chunk2doc_ranker_driver_min_ranker():
doc = create_document_to_score()
driver = SimpleChunk2DocRankDriver()
executor = MinRanker()
driver.attach(executor=executor, pea=None)
driver._apply_all(doc.chunks, doc)
assert len(doc.matches) == 4
assert doc.matches[0].id == 40
assert doc.matches[0].score.value == pytest.approx(1 / (1 + 4), 0.0001)
assert doc.matches[1].id == 50
assert doc.matches[1].score.value == pytest.approx(1 / (1 + 5), 0.0001)
assert doc.matches[2].id == 60
assert doc.matches[2].score.value == pytest.approx(1 / (1 + 6), 0.0001)
assert doc.matches[3].id == 70
assert doc.matches[3].score.value == pytest.approx(1 / (1 + 7), 0.0001)
for match in doc.matches:
# match score is computed w.r.t to doc.id
assert match.score.ref_id == doc.id
def test_chunk2doc_ranker_driver_MinRanker(self):
doc = create_document_to_score()
driver = SimpleChunk2DocRankDriver()
executor = MinRanker()
driver.attach(executor=executor, pea=None)
driver._apply_all(doc.chunks, doc)
assert len(doc.matches) == 4
assert doc.matches[0].id == 40
self.assertAlmostEqual(doc.matches[0].score.value, 1 / (1 + 4))
assert doc.matches[1].id == 50
self.assertAlmostEqual(doc.matches[1].score.value, 1 / (1 + 5))
assert doc.matches[2].id == 60
self.assertAlmostEqual(doc.matches[2].score.value, 1 / (1 + 6))
assert doc.matches[3].id == 70
self.assertAlmostEqual(doc.matches[3].score.value, 1 / (1 + 7))
for match in doc.matches:
# match score is computed w.r.t to doc.id
assert match.score.ref_id == doc.id