def test_combined_vector_sparse(self):
semspace = SemanticSpace(sparse_space, rows, columns, readme_title,
readme_desc, prenorm=False)
vector = semspace.combined_vector(['first', 'third'])
assert (space[[0,2], :].sum(0) == vector).all()
vector = semspace.combined_vector(['second', 'fourth'])
assert (space[[1,3], :].sum(0) == vector).all()
def test_metrics(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=False)
for metric in semspace.allowed_metrics():
print metric
pairs = [('first', 'second'), ('third', 'eighth'),
('twelfth', 'first'), ('twelfth', 'thirteenth')]
self.semspace.pair_distances(pairs, metric=metric)
def test_combined_vector_prenorm(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=True)
vector = semspace.combined_vector(['first', 'third'])
self.assertEqual(np.linalg.norm(vector), 1)
def test_prenorm_exception_on_non_cosine(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=True)
# cosine should work on prenormalized space
semspace.pair_distance('first', 'second', metric='cosine')
# but not euclidean
with self.assertRaises(Exception):
semspace.pair_distance('first', 'second', metric='euclidean')
def test_metrics(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=False)
for metric in semspace.allowed_metrics():
print(metric)
pairs = [('first', 'second'), ('third', 'eighth'),
('twelfth', 'first'), ('twelfth', 'thirteenth')]
self.semspace.pair_distances(pairs, metric=metric)
def test_combined_vector_sparse(self):
semspace = SemanticSpace(sparse_space,
rows,
columns,
readme_title,
readme_desc,
prenorm=False)
vector = semspace.combined_vector(['first', 'third'])
assert (space[[0, 2], :].sum(0) == vector).all()
vector = semspace.combined_vector(['second', 'fourth'])
assert (space[[1, 3], :].sum(0) == vector).all()
def load_semspace(semspace_path, semspace_format='semspace'):
"""Load a semantic space based on the path and format."""
global semspace
if semspace_format == 'ssmarket':
semspace = SemanticSpace.from_ssmarket(semspace_path,
prenorm=prenormalize)
return True
elif semspace_format == 'csv':
semspace = SemanticSpace.from_csv(semspace_path,
prenorm=prenormalize,
dtype=numpy_dtype)
return True
else:
raise Exception("Space format '%s' unknown!" % semspace_format)
def test_cosine_equals_prenorm(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=True)
cosine_non_prenorm = self.semspace.pair_distance('first', 'second',
metric='cosine')
cosine_prenorm = semspace.pair_distance('first', 'second',
metric='cosine')
self.assertEqual(cosine_non_prenorm, cosine_prenorm)
pairs = [('first', 'second'), ('third', 'eighth'),
('twelfth', 'first'), ('twelfth', 'thirteenth')]
pairs_sims_non_prenorm = self.semspace.pair_distances(pairs)
pairs_sims_prenorm = semspace.pair_distances(pairs)
self.assertEqual(pairs_sims_non_prenorm, pairs_sims_prenorm)
def test_prenormalization(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=True)
for row in semspace.vectors:
row_norm = np.linalg.norm(row)
print(row_norm, row)
self.assertAlmostEqual(row_norm, 1.0)
def test_cosine_equals_prenorm(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=True)
cosine_non_prenorm = self.semspace.pair_distance('first',
'second',
metric='cosine')
cosine_prenorm = semspace.pair_distance('first',
'second',
metric='cosine')
self.assertEqual(cosine_non_prenorm, cosine_prenorm)
pairs = [('first', 'second'), ('third', 'eighth'),
('twelfth', 'first'), ('twelfth', 'thirteenth')]
pairs_sims_non_prenorm = self.semspace.pair_distances(pairs)
pairs_sims_prenorm = semspace.pair_distances(pairs)
self.assertEqual(pairs_sims_non_prenorm, pairs_sims_prenorm)
def test_allowed_metrics_when_prenormed(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=False)
semspace_p = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=True)
self.assertIn('cosine', semspace.allowed_metrics())
self.assertIn('cosine', semspace_p.allowed_metrics())
self.assertNotIn('manhattan', semspace_p.allowed_metrics())
self.assertIn('manhattan', semspace.allowed_metrics())
def test_allowed_metrics_when_prenormed(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=False)
semspace_p = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=True)
self.assertIn('cosine', semspace.allowed_metrics())
self.assertIn('cosine', semspace_p.allowed_metrics())
self.assertNotIn('manhattan', semspace_p.allowed_metrics())
self.assertIn('manhattan', semspace.allowed_metrics())
def test_prenorm_exception_on_non_cosine(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=True)
# cosine should work on prenormalized space
semspace.pair_distance('first', 'second', metric='cosine')
# but not euclidean
with self.assertRaises(Exception):
semspace.pair_distance('first', 'second', metric='euclidean')
from semspaces.space import SemanticSpace
import numpy as np
space = np.array([[0.61502426, 0.35800892, 0.46591138],
[0.06256679, 0.80705953, 0.87805124],
[0.18189868, 0.37707662, 0.89973192],
[0.32667934, 0.0994168, 0.75457225],
[0.43300126, 0.17586539, 0.88097073],
[0.62085788, 0.29817756, 0.62991792],
[0.37163458, 0.86633926, 0.31679958],
[0.37416635, 0.82935107, 0.34275204],
[0.26996958, 0.57101081, 0.60706083],
[0.36690094, 0.70666147, 0.3300295],
[0.19479401, 0.3334173, 0.79296408]])
rows = [
'first', 'second', 'third', 'fourth', 'fifth', 'sixth', 'seventh',
'eighth', 'ninth', 'tenth', 'eleventh'
]
columns = ['one', 'two', 'three']
readme_title = 'Random semantic space'
readme_desc = 'Demo semantic space description.'
example_semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc)
def setUp(self):
self.semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc)
class TestSemanticSpace(unittest.TestCase):
def setUp(self):
self.semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc)
def test_defined_at_words(self):
assert self.semspace.defined_at('first')
assert not self.semspace.defined_at('twelfth')
assert not self.semspace.defined_at('one')
def test_defined_at_seqs(self):
assert self.semspace.defined_at(['first', 'second'])
assert not self.semspace.defined_at(['first', 'twelfth'])
assert not self.semspace.defined_at(['one', 'twelfth'])
def test_similarity_pairs(self):
self.assertAlmostEqual(self.semspace.pair_distance('first', 'first'),
0)
assert self.semspace.pair_distance('first', 'second') > 1e-10
def test_prenormalization(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=True)
for row in semspace.vectors:
row_norm = np.linalg.norm(row)
print(row_norm, row)
self.assertAlmostEqual(row_norm, 1.0)
def test_prenorm_exception_on_non_cosine(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=True)
# cosine should work on prenormalized space
semspace.pair_distance('first', 'second', metric='cosine')
# but not euclidean
with self.assertRaises(Exception):
semspace.pair_distance('first', 'second', metric='euclidean')
def test_cosine_equals_prenorm(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=True)
cosine_non_prenorm = self.semspace.pair_distance('first',
'second',
metric='cosine')
cosine_prenorm = semspace.pair_distance('first',
'second',
metric='cosine')
self.assertEqual(cosine_non_prenorm, cosine_prenorm)
pairs = [('first', 'second'), ('third', 'eighth'),
('twelfth', 'first'), ('twelfth', 'thirteenth')]
pairs_sims_non_prenorm = self.semspace.pair_distances(pairs)
pairs_sims_prenorm = semspace.pair_distances(pairs)
self.assertEqual(pairs_sims_non_prenorm, pairs_sims_prenorm)
def test_subset(self):
words = ['third', 'second', 'tenth', 'eighth']
subset = self.semspace.subset(words)
assert subset.vectors.shape == (4, 3)
assert not subset.defined_at('first')
assert subset.defined_at('second')
assert subset.defined_at('third')
assert subset.defined_at('eighth')
assert not subset.defined_at('ninth')
assert subset.defined_at('tenth')
self.assertAlmostEqual(self.semspace.pair_distance('second', 'third'),
subset.pair_distance('second', 'third'))
self.assertAlmostEqual(self.semspace.pair_distance('third', 'tenth'),
subset.pair_distance('third', 'tenth'))
def test_pair_distances(self):
pairs = [('first', 'second'), ('third', 'eighth'),
('twelfth', 'first'), ('twelfth', 'thirteenth')]
pairs_sims = self.semspace.pair_distances(pairs)
first_second = self.semspace.pair_distance('first', 'second')
assert pairs_sims[('first', 'second')] == first_second
assert ('third', 'eighth') in list(pairs_sims.keys())
assert ('twelfth', 'first') not in list(pairs_sims.keys())
assert ('twelfth', 'thirteenth') not in list(pairs_sims.keys())
pairs_sims_nan = self.semspace.pair_distances(pairs, na_val=True)
assert pairs_sims_nan[('first', 'second')] == first_second
assert ('third', 'eighth') in list(pairs_sims_nan.keys())
assert ('twelfth', 'first') in list(pairs_sims_nan.keys())
assert ('twelfth', 'thirteenth') in list(pairs_sims_nan.keys())
assert pairs_sims_nan[('twelfth', 'first')] is np.nan
assert pairs_sims_nan[('twelfth', 'thirteenth')] is np.nan
def test_vector_entropy(self):
self.assertAlmostEqual(self.semspace.vector_entropy('first'),
1.5502257500054266)
self.assertAlmostEqual(self.semspace.vector_entropy('fifth'),
1.3302170534376188)
self.assertAlmostEqual(self.semspace.vector_entropy('second'),
0.99871934706694587)
def test_allowed_metrics_when_prenormed(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=False)
semspace_p = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=True)
self.assertIn('cosine', semspace.allowed_metrics())
self.assertIn('cosine', semspace_p.allowed_metrics())
self.assertNotIn('manhattan', semspace_p.allowed_metrics())
self.assertIn('manhattan', semspace.allowed_metrics())
def test_metrics(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=False)
for metric in semspace.allowed_metrics():
print(metric)
pairs = [('first', 'second'), ('third', 'eighth'),
('twelfth', 'first'), ('twelfth', 'thirteenth')]
self.semspace.pair_distances(pairs, metric=metric)
def test_combined_vector(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=False)
vector = semspace.combined_vector(['first', 'third'])
assert (space[[0, 2], :].sum(0) == vector).all()
vector = semspace.combined_vector(['second', 'fourth'])
assert (space[[1, 3], :].sum(0) == vector).all()
def test_combined_vector_prenorm(self):
semspace = SemanticSpace(space,
rows,
columns,
readme_title,
readme_desc,
prenorm=True)
vector = semspace.combined_vector(['first', 'third'])
self.assertEqual(np.linalg.norm(vector), 1)
def test_combined_vector_sparse(self):
semspace = SemanticSpace(sparse_space,
rows,
columns,
readme_title,
readme_desc,
prenorm=False)
vector = semspace.combined_vector(['first', 'third'])
assert (space[[0, 2], :].sum(0) == vector).all()
vector = semspace.combined_vector(['second', 'fourth'])
assert (space[[1, 3], :].sum(0) == vector).all()
def test_combined_vector_prenorm(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=True)
vector = semspace.combined_vector(['first', 'third'])
self.assertEqual(np.linalg.norm(vector), 1)
class TestSemanticSpace(unittest.TestCase):
def setUp(self):
self.semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc)
def test_defined_at_words(self):
assert self.semspace.defined_at('first')
assert not self.semspace.defined_at('twelfth')
assert not self.semspace.defined_at('one')
def test_defined_at_seqs(self):
assert self.semspace.defined_at(['first', 'second'])
assert not self.semspace.defined_at(['first', 'twelfth'])
assert not self.semspace.defined_at(['one', 'twelfth'])
def test_similarity_pairs(self):
self.assertAlmostEqual(self.semspace.pair_distance('first', 'first'), 0)
assert self.semspace.pair_distance('first', 'second') > 1e-10
def test_prenormalization(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=True)
for row in semspace.vectors:
row_norm = np.linalg.norm(row)
print row_norm, row
self.assertAlmostEqual(row_norm, 1.0)
def test_prenorm_exception_on_non_cosine(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=True)
# cosine should work on prenormalized space
semspace.pair_distance('first', 'second', metric='cosine')
# but not euclidean
with self.assertRaises(Exception):
semspace.pair_distance('first', 'second', metric='euclidean')
def test_cosine_equals_prenorm(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=True)
cosine_non_prenorm = self.semspace.pair_distance('first', 'second',
metric='cosine')
cosine_prenorm = semspace.pair_distance('first', 'second',
metric='cosine')
self.assertEqual(cosine_non_prenorm, cosine_prenorm)
pairs = [('first', 'second'), ('third', 'eighth'),
('twelfth', 'first'), ('twelfth', 'thirteenth')]
pairs_sims_non_prenorm = self.semspace.pair_distances(pairs)
pairs_sims_prenorm = semspace.pair_distances(pairs)
self.assertEqual(pairs_sims_non_prenorm, pairs_sims_prenorm)
def test_subset(self):
words = ['third', 'second', 'tenth', 'eighth']
subset = self.semspace.subset(words)
assert subset.vectors.shape == (4, 3)
assert not subset.defined_at('first')
assert subset.defined_at('second')
assert subset.defined_at('third')
assert subset.defined_at('eighth')
assert not subset.defined_at('ninth')
assert subset.defined_at('tenth')
self.assertAlmostEqual(
self.semspace.pair_distance('second', 'third'),
subset.pair_distance('second', 'third'))
self.assertAlmostEqual(
self.semspace.pair_distance('third', 'tenth'),
subset.pair_distance('third', 'tenth'))
def test_pair_distances(self):
pairs = [('first', 'second'), ('third', 'eighth'),
('twelfth', 'first'), ('twelfth', 'thirteenth')]
pairs_sims = self.semspace.pair_distances(pairs)
first_second = self.semspace.pair_distance('first', 'second')
assert pairs_sims[('first', 'second')] == first_second
assert ('third', 'eighth') in pairs_sims.keys()
assert ('twelfth', 'first') not in pairs_sims.keys()
assert ('twelfth', 'thirteenth') not in pairs_sims.keys()
pairs_sims_nan = self.semspace.pair_distances(pairs, na_val=True)
assert pairs_sims_nan[('first', 'second')] == first_second
assert ('third', 'eighth') in pairs_sims_nan.keys()
assert ('twelfth', 'first') in pairs_sims_nan.keys()
assert ('twelfth', 'thirteenth') in pairs_sims_nan.keys()
assert pairs_sims_nan[('twelfth', 'first')] is np.nan
assert pairs_sims_nan[('twelfth', 'thirteenth')] is np.nan
def test_vector_entropy(self):
self.assertAlmostEqual(
self.semspace.vector_entropy('first'),
1.5502257500054266)
self.assertAlmostEqual(
self.semspace.vector_entropy('fifth'),
1.3302170534376188)
self.assertAlmostEqual(
self.semspace.vector_entropy('second'),
0.99871934706694587)
def test_allowed_metrics_when_prenormed(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=False)
semspace_p = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=True)
self.assertIn('cosine', semspace.allowed_metrics())
self.assertIn('cosine', semspace_p.allowed_metrics())
self.assertNotIn('manhattan', semspace_p.allowed_metrics())
self.assertIn('manhattan', semspace.allowed_metrics())
def test_metrics(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=False)
for metric in semspace.allowed_metrics():
print metric
pairs = [('first', 'second'), ('third', 'eighth'),
('twelfth', 'first'), ('twelfth', 'thirteenth')]
self.semspace.pair_distances(pairs, metric=metric)
def test_combined_vector(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=False)
vector = semspace.combined_vector(['first', 'third'])
assert (space[[0,2], :].sum(0) == vector).all()
vector = semspace.combined_vector(['second', 'fourth'])
assert (space[[1,3], :].sum(0) == vector).all()
def test_combined_vector_prenorm(self):
semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc, prenorm=True)
vector = semspace.combined_vector(['first', 'third'])
self.assertEqual(np.linalg.norm(vector), 1)
def test_combined_vector_sparse(self):
semspace = SemanticSpace(sparse_space, rows, columns, readme_title,
readme_desc, prenorm=False)
vector = semspace.combined_vector(['first', 'third'])
assert (space[[0,2], :].sum(0) == vector).all()
vector = semspace.combined_vector(['second', 'fourth'])
assert (space[[1,3], :].sum(0) == vector).all()
def setUp(self):
self.semspace = SemanticSpace(space, rows, columns, readme_title,
readme_desc)
random_baseline = False
sample_vocab = True
data_dir = os.path.join(os.getcwd(), 'data/')
out_dir = os.path.join(os.getcwd(), 'output/')
if not os.path.exists(out_dir):
os.makedirs(out_dir)
# LOAD DATA
print(
datetime.now().strftime("%d/%m/%Y %H:%M:%S: Started loading the data..."))
embedding_space = SemanticSpace.from_csv(os.path.join(
data_dir,
'/home/gcassani/Resources/Embeddings/embedding_space.cbow.ukwac.subtlex.300dims.w5.w2v'
),
prenorm=True)
w2v_words = embedding_space.included_words()
celex = json.load(open(os.path.join(data_dir, 'celex_dict.json')))
aoa_words, aoa_norms = aoa.read(os.path.join(data_dir, "AoA.xlsx"))
w2aoa = pd.Series(aoa_norms["Rating.Mean"].values,
index=aoa_norms["Word"]).to_dict()
w2concr = concreteness.read(os.path.join(data_dir, "concreteness.txt"))
w2val = valence.read(os.path.join(data_dir, "valence.csv"))
w2freq = subtlex.read(os.path.join(data_dir, "subtlex.csv"))
w2old = old20.read(os.path.join(data_dir, "word2old.csv"))
mono = list(morpholex.read_mono(os.path.join(data_dir, "MorphoLEX_en.xlsx")))
def load_space():
print('Loading semantic space...')
return SemanticSpace.from_csv('spaces/lemmas.w2v.gz')