def get_matches(evaluation_references, sample_N):
# Take a random sample of the evaluation references to find matches for
match_data = evaluation_references.sample(n=sample_N * 2,
random_state=0).reset_index()
match_data['Title'] = match_data['title']
match_data_positive = match_data.iloc[0:sample_N]
match_data_negative = match_data.iloc[sample_N:]
evaluation_references_without_negative = evaluation_references.loc[
~evaluation_references['uber_id'].isin(match_data_negative['uber_id'])]
fuzzy_matcher = FuzzyMatcher(evaluation_references_without_negative, -1)
match_data_pos_neg = pd.concat([match_data_positive, match_data_negative])
eval_references = []
for i, ref in match_data_pos_neg.iterrows():
match_ref = fuzzy_matcher.match(ref)
eval_references.append(match_ref)
eval_references = pd.DataFrame(eval_references)
eval_references["Title Length"] = [
len(title) for title in eval_references["Extracted title"]
]
eval_references["Match Type"] = ["Positive"] * sample_N + ["Negative"
] * sample_N
return eval_references
def test_close_match(self):
real_publications = pd.DataFrame({
'title': ['Malaria is caused by mosquitoes'],
'uber_id': [1]
})
threshold = settings.FUZZYMATCH_SIMILARITY_THRESHOLD
fuzzy_matcher = FuzzyMatcher(real_publications, threshold)
reference = {
'Document id': 10,
'Reference id': 11,
'Title': 'Malaria'
}
matched_publication = fuzzy_matcher.match(reference)
self.assertEqual(matched_publication, None)
def test_no_string_titles(self):
real_publications = pd.DataFrame({
'title': [1,2]
})
threshold = 75
with self.assertRaises(AttributeError):
FuzzyMatcher(real_publications, threshold)
def test_empty_title(self):
real_publications = pd.DataFrame({
'title': []
})
threshold = 75
with self.assertRaises(ValueError):
FuzzyMatcher(real_publications, threshold)
def evaluate_match_references(evaluation_references, match_threshold,
length_threshold, sample_N):
# Take a random sample of the evaluation references to find matches for
match_data = evaluation_references.sample(n=sample_N * 2,
random_state=0).reset_index()
match_data['Title'] = match_data['title']
match_data_positive = match_data.iloc[0:sample_N]
match_data_negative = match_data.iloc[sample_N:]
evaluation_references_without_negative = evaluation_references.loc[
~evaluation_references['uber_id'].isin(match_data_negative['uber_id'])]
fuzzy_matcher = FuzzyMatcher(evaluation_references_without_negative,
match_threshold, length_threshold)
predictions = predict_match_data(
match_data=match_data_positive.to_dict('records') +
match_data_negative.to_dict('records'),
matcher=fuzzy_matcher)
actual = match_data_positive['Reference id'].to_list() + [None] * sample_N
metrics = evaluate_metric(actual, predictions)
return metrics
def init_fuzzy_matcher(self):
real_publications = pd.DataFrame({
'title': ['Malaria', 'Zika'],
'uber_id': [1, 2]
})
threshold = settings.FUZZYMATCH_SIMILARITY_THRESHOLD
fuzzy_matcher = FuzzyMatcher(real_publications, threshold)
return fuzzy_matcher
def test_init_variables(self):
real_publications = pd.DataFrame({
'title': ['Malaria', 'Zika']
})
threshold = 0
fuzzy_matcher = FuzzyMatcher(real_publications, threshold)
assert_frame_equal(
fuzzy_matcher.publications, real_publications
)
self.assertEqual(
fuzzy_matcher.similarity_threshold, threshold
)
self.assertTrue(
fuzzy_matcher.tfidf_matrix.size != 0
)
def test_no_title(self):
real_publications = pd.DataFrame({})
threshold = 75
with self.assertRaises(KeyError):
FuzzyMatcher(real_publications, threshold)