def test_vector_space_wrapper_filter():
wrap = VectorSpaceWrapper(frame=TEST_FRAME)
wrap.load()
ok_('/c/pl/kombinacja' in wrap.similar_terms('/c/en/nordic_combined', filter='/c/pl',
limit=1).index)
ok_('/c/en/present' in wrap.similar_terms('/c/en/gift', filter='/c/en/present', limit=1).index)
def test_missing_language():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
# The frame contains no Esperanto, of course, so the out-of-vocabulary
# mechanism will fail. We should simply get no results, not crash.
similarity = vectors.similar_terms('/c/eo/ekzemplo')
eq_(len(similarity), 0)
def test_similar_terms_filter():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
ok_('/c/pl/kombinacja' in vectors.similar_terms(
'/c/en/nordic_combined', filter='/c/pl', limit=1).index)
ok_('/c/en/present' in vectors.similar_terms(
'/c/en/gift', filter='/c/en/present', limit=1).index)
def test_missing_language():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
# The frame contains no Esperanto, of course, so the out-of-vocabulary
# mechanism will fail. We should simply get no results, not crash.
similarity = vectors.similar_terms('/c/eo/ekzemplo')
eq_(len(similarity), 0)
def test_match_prefix():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
term = '/c/en/figure_skate'
expected_prefix_matches = [('/c/en/figure', 0.0033333333333333335),
('/c/en/figure skater', 0.0033333333333333335),
('/c/en/figure skating', 0.0033333333333333335)]
prefix_matches = vectors._match_prefix(term=term, prefix_weight=0.01)
eq_(expected_prefix_matches, prefix_matches)
def test_cache_with_oov():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
# check the vector of all zeros is returned if the term is not present
ok_(not vectors.get_vector('/c/en/test', oov_vector=False).any())
# If include_neighbors=True, the neighbor of 'test' in ConceptNet ('trial')
# will be used to approximate its vector
ok_(vectors.get_vector('/c/en/test', oov_vector=True).any())
def test_cache_with_oov():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
# check the vector of all zeros is returned if the term is not present
ok_(not vectors.get_vector('/c/en/test', oov_vector=False).any())
# If include_neighbors=True, the neighbor of 'test' in ConceptNet ('trial')
# will be used to approximate its vector
ok_(vectors.get_vector('/c/en/test', oov_vector=True).any())
def measure_bias(frame):
"""
Return a DataFrame that measures biases in a semantic space, on four
data sets:
- Gender
- Fine-grained ethnicity
- Coarse-grained ethnicity
- Religious beliefs
"""
vsw = VectorSpaceWrapper(frame=frame)
vsw.load()
gender_binary_axis = normalize_vec(
get_category_axis(frame, FEMALE_WORDS) -
get_category_axis(frame, MALE_WORDS))
gender_bias_numbers = []
for female_biased_word, male_biased_word in GENDER_BIAS_PAIRS:
female_biased_uri = standardized_uri('en', female_biased_word)
male_biased_uri = standardized_uri('en', male_biased_word)
diff = normalize_vec(
vsw.get_vector(female_biased_uri) -
vsw.get_vector(male_biased_uri)).dot(gender_binary_axis)
gender_bias_numbers.append(diff)
mean = np.mean(gender_bias_numbers)
sem = scipy.stats.sem(gender_bias_numbers)
gender_bias = pd.Series([mean, mean - sem * 2, mean + sem * 2],
index=['bias', 'low', 'high'])
stereotype_vecs_1 = get_vocabulary_vectors(frame, PEOPLE_BY_ETHNICITY)
stereotype_vecs_2 = get_vocabulary_vectors(frame, ETHNIC_STEREOTYPE_TERMS)
fine_ethnic_bias = correlation_bias(stereotype_vecs_1, stereotype_vecs_2)
stereotype_vecs_1 = get_vocabulary_vectors(frame, COARSE_ETHNICITY_TERMS)
stereotype_vecs_2 = get_vocabulary_vectors(frame, ETHNIC_STEREOTYPE_TERMS)
coarse_ethnic_bias = correlation_bias(stereotype_vecs_1, stereotype_vecs_2)
stereotype_vecs_1 = pd.DataFrame(
np.vstack(
[get_category_axis(frame, names) for names in ETHNIC_NAME_SETS]))
stereotype_vecs_2 = get_vocabulary_vectors(frame, ETHNIC_STEREOTYPE_TERMS)
name_ethnic_bias = correlation_bias(stereotype_vecs_1, stereotype_vecs_2)
stereotype_vecs_1 = get_vocabulary_vectors(frame, PEOPLE_BY_BELIEF)
stereotype_vecs_2 = get_vocabulary_vectors(frame, BELIEF_STEREOTYPE_TERMS)
belief_bias = correlation_bias(stereotype_vecs_1, stereotype_vecs_2)
return pd.DataFrame({
'gender': gender_bias,
'ethnicity-fine': fine_ethnic_bias,
'ethnicity-coarse': coarse_ethnic_bias,
'ethnicity-names': name_ethnic_bias,
'beliefs': belief_bias
}).T
def test_expand_terms(multi_ling_frame):
vectors = VectorSpaceWrapper(frame=multi_ling_frame)
vectors.load()
term = [('/c/en/ski_jumper', 1.0)]
expanded_terms = vectors.expand_terms(terms=term, oov_vector=True)
expected_expanded_terms = [
('/c/en/ski_jumper', 0.9900990099009901),
('/c/en/ski_jumping', 0.009900990099009901),
]
assert expected_expanded_terms == expanded_terms
def test_match_prefix():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
term = '/c/en/figure_skate'
expected_prefix_matches = [
('/c/en/figure', 0.0033333333333333335),
('/c/en/figure skater', 0.0033333333333333335),
('/c/en/figure skating', 0.0033333333333333335),
]
prefix_matches = vectors._match_prefix(term=term, prefix_weight=0.01)
eq_(expected_prefix_matches, prefix_matches)
def test_similar_terms_filter(multi_ling_frame):
vectors = VectorSpaceWrapper(frame=multi_ling_frame)
vectors.load()
assert ('/c/pl/kombinacja'
in vectors.similar_terms('/c/en/nordic_combined',
filter='/c/pl',
limit=1).index)
assert ('/c/en/present' in vectors.similar_terms('/c/en/gift',
filter='/c/en/present',
limit=1).index)
def test_match_prefix(simple_frame):
vectors = VectorSpaceWrapper(frame=simple_frame)
vectors.load()
term = '/c/en/figure_skate'
expected_prefix_matches = [
('/c/en/figure', 0.0033333333333333335),
('/c/en/figure skater', 0.0033333333333333335),
('/c/en/figure skating', 0.0033333333333333335),
]
prefix_matches = vectors._match_prefix(term=term, prefix_weight=0.01)
assert expected_prefix_matches == prefix_matches
def test_similar_terms():
"""
Check if VectorSpaceWrapper's index is sorted and its elements are concepts.
"""
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
ok_('/c/en/figure skating' in vectors.similar_terms('/c/en/figure skating',
limit=3).index)
ok_('/c/en/figure skater' in vectors.similar_terms('/c/en/figure skating',
limit=3).index)
ok_('/c/en/figure' in vectors.similar_terms('/c/en/figure skating',
limit=3).index)
def test_lookup_neighbors():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
term = '/c/pl/skoki_narciarskie'
neighbors = vectors._find_neighbors(term=term,
limit_per_term=10,
weight=1.0)
expected_neighbors = [('/c/en/ski_jumping', 0.02),
('http://pl.dbpedia.org/resource/Skoki_narciarskie',
0.01), ('/c/en/ski_jumping', 0.01),
('/c/en/ski_jumping', 0.005)]
eq_(expected_neighbors, neighbors)
def test_similar_terms_filter():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
ok_(
'/c/pl/kombinacja'
in vectors.similar_terms('/c/en/nordic_combined', filter='/c/pl', limit=1).index
)
ok_(
'/c/en/present'
in vectors.similar_terms('/c/en/gift', filter='/c/en/present', limit=1).index
)
def test_similar_terms(simple_frame):
"""
Check if VectorSpaceWrapper's index is sorted and its elements are concepts.
"""
vectors = VectorSpaceWrapper(frame=simple_frame)
vectors.load()
assert ('/c/en/figure skating'
in vectors.similar_terms('/c/en/figure skating', limit=3).index)
assert ('/c/en/figure skater'
in vectors.similar_terms('/c/en/figure skating', limit=3).index)
assert ('/c/en/figure' in vectors.similar_terms('/c/en/figure skating',
limit=3).index)
def test_lookup_neighbors():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
term = '/c/pl/skoki_narciarskie'
neighbors = vectors._find_neighbors(term=term, limit_per_term=10, weight=1.0)
expected_neighbors = {
('/c/en/ski_jumping', 0.02),
('/c/en/ski_jumping', 0.01),
('http://pl.dbpedia.org/resource/Skoki_narciarskie', 0.01),
('/c/de/skispringen', 0.01),
('/c/en/ski_jumping', 0.005),
}
eq_(expected_neighbors, set(neighbors))
def test_expand_terms():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
term = [('/c/en/ski_jumper', 1.0)]
expanded_terms = vectors.expand_terms(terms=term, limit_per_term=2, oov_vector=True)
expected_expanded_terms = [
('/c/en/ski_jumper', 0.9523809523809523),
('/c/pt/saltadores_de_esqui', 0.019047619047619046),
('/c/pl/skoczek_narciarski', 0.019047619047619046),
('/c/en/ski_jumping', 0.009523809523809523),
]
eq_(expected_expanded_terms, expanded_terms)
def test_load():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
ok_(vectors.frame is not None)
ok_(vectors.small_frame is not None)
ok_(all(label.startswith('/c/en/') for label in vectors.frame.index))
ok_(vectors.frame.index.is_monotonic_increasing)
ok_(vectors.small_frame.shape[1] <= 100)
ok_(vectors._trie is not None)
# test there are no transformations to raw terms other than adding the
# English tag
ok_('/c/en/figure skater' in vectors.frame.index) # no underscore
ok_('/c/en/Island' in vectors.frame.index) # no case folding
def test_load():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
ok_(vectors.frame is not None)
ok_(vectors.small_frame is not None)
ok_(all(label.startswith('/c/en/') for label in vectors.frame.index))
ok_(vectors.frame.index.is_monotonic_increasing)
ok_(vectors.small_frame.shape[1] <= 100)
ok_(vectors._trie is not None)
# test there are no transformations to raw terms other than adding the
# English tag
ok_('/c/en/figure skater' in vectors.frame.index) # no underscore
ok_('/c/en/Island' in vectors.frame.index) # no case folding
def test_load(simple_frame):
vectors = VectorSpaceWrapper(frame=simple_frame)
vectors.load()
assert vectors.frame is not None
assert vectors.small_frame is not None
assert all(label.startswith('/c/en/') for label in vectors.frame.index)
assert vectors.frame.index.is_monotonic_increasing
assert vectors.small_frame.shape[1] <= 100
assert vectors._trie is not None
# test there are no transformations to raw terms other than adding the
# English tag
assert '/c/en/figure skater' in vectors.frame.index # no underscore
assert '/c/en/Island' in vectors.frame.index # no case folding
def test_expand_terms():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
term = [('/c/en/ski_jumper', 1.0)]
expanded_terms = vectors.expand_terms(terms=term,
limit_per_term=2,
oov_vector=True)
# /c/en/bounder and /c/en/skier from neighbor search
# /c/en/ski_jumping from prefix match
expected_expanded_terms = [('/c/en/ski_jumper', 0.9523809523809523),
('/c/en/bounder', 0.019047619047619046),
('/c/en/skier', 0.019047619047619046),
('/c/en/ski_jumping', 0.009523809523809523)]
eq_(expected_expanded_terms, expanded_terms)
def test_expand_terms():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
term = [('/c/en/ski_jumper', 1.0)]
expanded_terms = vectors.expand_terms(terms=term,
limit_per_term=2,
oov_vector=True)
expected_expanded_terms = [
('/c/en/ski_jumper', 0.9523809523809523),
('/c/pt/saltadores_de_esqui', 0.019047619047619046),
('/c/pl/skoczek_narciarski', 0.019047619047619046),
('/c/en/ski_jumping', 0.009523809523809523),
]
eq_(expected_expanded_terms, expanded_terms)
def test_similar_terms():
"""
Check if VectorSpaceWrapper's index is sorted and its elements are concepts.
"""
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
ok_(
'/c/en/figure skating'
in vectors.similar_terms('/c/en/figure skating', limit=3).index
)
ok_(
'/c/en/figure skater'
in vectors.similar_terms('/c/en/figure skating', limit=3).index
)
ok_('/c/en/figure' in vectors.similar_terms('/c/en/figure skating', limit=3).index)
def test_vector_space_wrapper():
"""
Check if VectorSpaceWrapper's index is sorted and its elements are concepts.
"""
wrap = VectorSpaceWrapper(frame=TEST_FRAME)
wrap.load()
ok_(all(is_term(label) for label in wrap.frame.index))
ok_(wrap.frame.index.is_monotonic_increasing)
# test there are no transformations to raw terms other than adding the english tag
ok_('/c/en/figure skater' in wrap.frame.index) # no underscore
ok_('/c/en/Island' in wrap.frame.index) # no case folding
# test index_prefix_range
ok_(wrap.index_prefix_range('/c/en/figure') == (3, 6))
ok_(wrap.index_prefix_range('/c/en/skating') == (0, 0))
# test_similar_terms
ok_('/c/en/figure skating' in wrap.similar_terms('/c/en/figure skating', limit=3).index)
ok_('/c/en/figure skater' in wrap.similar_terms('/c/en/figure skating', limit=3).index)
ok_('/c/en/figure' in wrap.similar_terms('/c/en/figure skating', limit=3).index)
def test_vector_space_wrapper(frame=None):
"""
Check if VectorSpaceWrapper's index is sorted and its elements are concepts.
"""
# Load a VSW from a user-supplied frame
if frame:
frame = load_any_embeddings(frame)
wrap = VectorSpaceWrapper(frame=frame)
wrap.load()
ok_(all(label.startswith('/c') for label in wrap.frame.index[1:]))
ok_(wrap.frame.index.is_monotonic_increasing)
# Load a VSW from a filename
vector_filename = DATA + '/vectors/glove12-840B.h5'
wrap = VectorSpaceWrapper(vector_filename=vector_filename)
wrap.load()
ok_(all(label.startswith('/c') for label in wrap.frame.index[1:]))
ok_(wrap.frame.index.is_monotonic_increasing)
# Load a VSW from a frame
frame = load_any_embeddings(DATA + '/vectors/glove12-840B.h5')
wrap = VectorSpaceWrapper(frame=frame)
wrap.load()
ok_(all(label.startswith('/c') for label in wrap.frame.index[1:]))
ok_(wrap.frame.index.is_monotonic_increasing)
def test_index_prefix_range():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
eq_(vectors._index_prefix_range('/c/en/figure'), (3, 6))
eq_(vectors._index_prefix_range('/c/en/skating'), (0, 0))
def test_index_prefix_range():
vectors = VectorSpaceWrapper(frame=TEST_FRAME)
vectors.load()
eq_(vectors._index_prefix_range('/c/en/figure'), (3, 6))
eq_(vectors._index_prefix_range('/c/en/skating'), (0, 0))
def test_cache_with_oov(multi_ling_frame):
vectors = VectorSpaceWrapper(frame=multi_ling_frame)
vectors.load()
# check the vector of all zeros is returned if the term is not present
assert not vectors.get_vector('/c/en/test', oov_vector=False).any()
def test_index_prefix_range(simple_frame):
vectors = VectorSpaceWrapper(frame=simple_frame)
vectors.load()
assert vectors._index_prefix_range('/c/en/figure') == (3, 6)
assert vectors._index_prefix_range('/c/en/skating') == (0, 0)
