def test_feat_individual_weighted_cos():
generator = FeatureGenerator(DATA_SCORING, CLARQS_SCORING)
p_id = 0
q_ids = [1, 2, 3, 4]
q_scores = [10, 2, 1, 1]
score = generator.feat_individual_cos_weighted(p_id, q_ids, q_scores)
assert score == pytest.approx(10.07, rel=1e-2)
def test_feat_individual_cos():
generator = FeatureGenerator(DATA_SCORING, CLARQS_SCORING)
p_id = 0
q_ids = [1, 2, 3, 4]
q_scores = [1, 1, 1, 1]
assert generator.feat_individual_cos(p_id, q_ids,
q_scores) == pytest.approx(2.707,
rel=1e-3)
def test_feat_post_contains_blockquote():
post1 = "this is first >line"
post2 = "this is my question?\n> some >test"
data = _two_post_df(post1, post2)
generator = FeatureGenerator(data, df_clarqs=None)
assert generator.feat_post_contains_blockquote(
p_id=0, q_ids=[], q_scores=[]) is False
assert generator.feat_post_contains_blockquote(
p_id=1, q_ids=[], q_scores=[]) is True
def test_feat_post_contains_preformatted():
post1 = "this is first line\n code\n more code\n more code\nno code"
post2 = "this is first line\nand post has no code in it"
data = _two_post_df(post1, post2)
generator = FeatureGenerator(data, df_clarqs=None)
assert generator.feat_post_contains_preformatted(
p_id=0, q_ids=[], q_scores=[]) is True
assert generator.feat_post_contains_preformatted(
p_id=1, q_ids=[], q_scores=[]) is False
def test_feat_post_length():
data = pd.DataFrame([
[0, 'title', 'body1 body2 body3', 'tags', 1],
],
columns=['id', 'title', 'body', 'tags', 'label'])
data.set_index('id', inplace=True)
generator = FeatureGenerator(data, df_clarqs=None)
length = generator.feat_post_length(p_id=0, q_ids=[], q_scores=[])
assert length == 5
def test_feat_fraction():
generator = FeatureGenerator(DATA, CLARQS)
proportion = generator.feat_fraction(0, [7, 0, 1, 2, 3],
[100, 3, 2, 1, 0.5])
assert proportion == 0.2
proportion = generator.feat_fraction(0, [5], [100, 3, 2, 1, 0.5])
assert proportion == 1
proportion = generator.feat_fraction(0, [0], [100, 3, 2, 1, 0.5])
assert proportion == 0
def test_score_cosine():
p_vec = [2, 1, 0]
cq_vec = [2, 1, 2]
assert FeatureGenerator.score_cosine(p_vec,
cq_vec) == pytest.approx(0.745,
rel=1e-3)
def test_vectorize_subjects():
# tokenized post and clarification questions
p = ['t1', 't1', 't2']
cq1 = ['t4', 't1']
cq2 = ['t1']
cq3 = ['t2']
cq4 = ['t4']
cq = [cq1, cq2, cq3, cq4]
p_vec, cq_vec = FeatureGenerator.vectorize_subjects(p, cq)
assert p_vec == [2, 1, 0]
assert cq_vec == [2, 1, 2]
def test_feat_ratio():
generator = FeatureGenerator(DATA, CLARQS)
ratio = generator.feat_ratio(0, [0, 1, 5, 6], [100, 3, 2, 1, 0.5])
assert ratio == 1
ratio = generator.feat_ratio(0, [5, 6, 7, 0, 1], [100, 3, 2, 1, 0.5])
assert ratio == 1.5
ratio = generator.feat_ratio(0, [5, 6, 0, 1, 2, 3], [100, 3, 2, 1, 0.5])
assert ratio == 0.5
ratio = generator.feat_ratio(0, [0], [100, 3, 2, 1, 0.5])
assert ratio == 0
# return number of clear if there are no unclear similar questions
ratio = generator.feat_ratio(0, [6, 7, 8], [100, 3, 2, 1, 0.5])
assert ratio == 3
def test_feat_majority():
generator = FeatureGenerator(DATA, CLARQS)
majority = generator.feat_majority(0, [0, 1, 4, 5, 6], [100, 3, 2, 1, 0.5])
assert majority == 1
majority = generator.feat_majority(0, [0, 1, 5, 6, 7], [100, 3, 2, 1, 0.5])
assert majority == 0
majority = generator.feat_majority(0, [0, 1, 4, 5, 6, 7],
[100, 3, 2, 1, 0.5])
assert majority == 0
majority = generator.feat_majority(0, [5, 6, 7, 0, 1, 4],
[100, 3, 2, 1, 0.5])
assert majority == 0
def test_compute_paper_example():
title = 'Simplest XML editor'
body = "I need the simplest editor with utf8 support for editing xml files; It's for a non programmer (so no atom or the like), to edit existing files. Any suggestion?"
tags = '<xml><utf8><editors>'
data = pd.DataFrame([
[0, title, body, tags, None],
[1, 'title', 'body', 'tags', 1],
[2, 'title', 'body', 'tags', 0],
[3, 'title', 'body', 'tags', 1],
],
columns=['id', 'title', 'body', 'tags', 'label'])
data.set_index('id', inplace=True)
clarqs = pd.DataFrame(
[[1, "What operating system?"], [3, "Have you tried atom?"]],
columns=['id', 'clarification_question'])
clarqs.set_index('id', inplace=True)
generator = FeatureGenerator(data, clarqs)
p_id = 0
q_ids = [1, 2, 3]
q_scores = [5, 2, 1]
def blank(p_id, q_ids, q_scores):
return ''
feature_table = [
('(i) Features on q', blank),
('==================', blank),
('Len(q)', generator.feat_post_length),
('ContainsPre(q)', generator.feat_post_contains_preformatted),
('ContainsQuote(q)', generator.feat_post_contains_blockquote),
('ContainsQuest(q)', generator.feat_post_contains_questionmark),
('Readability(q)', generator.feat_post_readability),
('(ii) Features on Q\'', blank),
('==================', blank),
('SimSum(q,Q\')', generator.feat_sim_sum),
('SimMax(q,Q\')', generator.feat_sim_max),
('SimAvg(q,Q\')', generator.feat_sim_avg),
('LenSim(Q\')', generator.feat_num_similar),
('SimUnclear(Q\')', generator.feat_num_unclear),
('SimClear(Q\')', generator.feat_num_clear),
(
'Majority(Q\')',
generator.feat_majority,
),
('Ratio(Q\')', generator.feat_ratio),
('Fraction(Q\')', generator.feat_fraction),
('(ii) Features on CQ\'', blank),
('==================', blank),
]
features_unclear = [
('CQGlobal(q,CQ\')', generator.feat_global_cos),
('CQIndividual(q,CQ\')', generator.feat_individual_cos),
('CQWeighted(q,CQ\')', generator.feat_individual_cos_weighted)
]
print()
for name, feature in feature_table:
print('{} = {}'.format(name, feature(p_id, q_ids, q_scores)))
for name, feature in features_unclear:
print('{} = {}'.format(name,
feature(p_id, q_ids=[1, 3], q_scores=[5, 1])))
def test_feat_num_similar():
num_similar = FeatureGenerator.feat_num_similar(p_id=None,
q_ids=[1, 2, 3, 4, 5],
q_scores=[5, 4, 3, 2, 1])
assert num_similar == 5
def test_feat_sim_avg():
sim_avg = FeatureGenerator.feat_sim_avg(p_id=None,
q_ids=[],
q_scores=[5, 4, 3, 2, 1])
assert sim_avg == 3
def test_feat_sim_max():
sim_max = FeatureGenerator.feat_sim_max(p_id=None,
q_ids=[],
q_scores=[5, 4, 3, 2, 1])
assert sim_max == 5
def test_feat_sim_sum():
sim_sum = FeatureGenerator.feat_sim_sum(p_id=None,
q_ids=[],
q_scores=[5, 4, 3, 2, 1])
assert sim_sum == 15