def __maxSimC(self, vecs_a, context_a, vecs_b, context_b):
vecmanip = VectorManipulation()
closest_a = self.__closestSenseContext(vecs_a, context_a)
closest_b = self.__closestSenseContext(vecs_b, context_b)
result = vecmanip.cosine_similarity(closest_a, closest_b)
return result
def globalSim(self, vecs_a, vecs_b):
vecmanip = VectorManipulation()
if not vecs_a or not vecs_b:
global_sim = 0.0
else:
global_a = numpy.average(vecs_a, axis=0)
global_b = numpy.average(vecs_b, axis=0)
global_sim = vecmanip.cosine_similarity(global_a, global_b)
return global_sim
def maxSim(self, vecs_a, vecs_b):
vecmanip = VectorManipulation()
highest = -1.0
for vec_a in vecs_a:
for vec_b in vecs_b:
tmp_high = vecmanip.cosine_similarity(vec_a, vec_b)
if tmp_high > highest:
highest = tmp_high
return highest
def test_SpearmanGlobal(self):
fio = FileManipulation()
vec_manip = VectorManipulation()
gold_tokens = fio.readFileLine('scws.txt')
ruby_global = fio.readFileLine('scws_globsim.txt')
ruby_globalc = fio.readFileLine('scws_globsimc.txt')
t11, __ = vec_manip.spearmanCorrelation(gold_tokens,
ruby_global) # not using rho
t21, __ = vec_manip.spearmanCorrelation(gold_tokens,
ruby_globalc) # not using rho
self.assertEqual('0.6670118503142607', str(t11))
self.assertEqual(numpy.float64('0.2969117412433547'), t21)
def test_SpearmanMax(self):
fio = FileManipulation()
vec_manip = VectorManipulation()
gold_tokens = fio.readFileLine('scws.txt')
ruby_max = fio.readFileLine('scws_maxsim.txt')
ruby_maxc = fio.readFileLine('scws_maxsimc.txt')
t11, __ = vec_manip.spearmanCorrelation(gold_tokens,
ruby_max) # not using rho
t21, __ = vec_manip.spearmanCorrelation(gold_tokens,
ruby_maxc) # not using rho
self.assertEqual('0.6127420529962664', str(t11))
self.assertEqual(numpy.float64('0.6367583108796157'), t21)
def test_SpearmanAvg(self):
fio = FileManipulation()
vec_manip = VectorManipulation()
gold_tokens = fio.readFileLine('scws.txt')
ruby_avg = fio.readFileLine('scws_avgsim.txt')
ruby_avgc = fio.readFileLine('scws_avgsimc.txt')
t11, __ = vec_manip.spearmanCorrelation(gold_tokens,
ruby_avg) # not using rho
t21, __ = vec_manip.spearmanCorrelation(gold_tokens,
ruby_avgc) # not using rho
self.assertEqual('0.6672948584312471', str(t11))
self.assertEqual(numpy.float64('0.5809138966365319'), t21)
def __closestSenseContext(self, synset_vecs, contextvec):
vecmanip = VectorManipulation()
high_so_far = -1.0
nearest = []
for synset_vec in synset_vecs: # closest sense (synset_vec) of 'word-A' to its context
context_sim = vecmanip.cosine_similarity(synset_vec, contextvec)
if context_sim > high_so_far:
high_so_far = context_sim
nearest = synset_vec
return nearest
def avgSim(self, vecs_a, vecs_b):
vecmanip = VectorManipulation()
partial_sim = 0.0
for vec_a in vecs_a:
for vec_b in vecs_b:
tmp_ab = vecmanip.cosine_similarity(vec_a, vec_b)
partial_sim += tmp_ab
if not vecs_a or not vecs_b:
final_sim = 0.0
else:
final_sim = (partial_sim / (len(vecs_a) * len(vecs_b)))
return final_sim
def __avgSimC(self, vecs_a, context_a, vecs_b, context_b):
vecmanip = VectorManipulation()
partial_sim = 0.0
for vec_a in vecs_a:
pcwa = vecmanip.cosine_similarity(vec_a, context_a)
for vec_b in vecs_b:
pcwb = vecmanip.cosine_similarity(vec_b, context_b)
dwab = vecmanip.cosine_similarity(vec_a, vec_b)
partial_sim += pcwa * pcwb * dwab
if not vecs_a or not vecs_b:
final_sim = 0.0
else:
final_sim = (partial_sim / (len(vecs_a) * len(vecs_b)))
return final_sim
def __contextParser(self, text_items, trained_model):
track_synset = SynsetParserVector()
vector_manip = VectorManipulation()
context_vector = []
for text_item in text_items:
synsets = wn.synsets(text_item)
for synset in synsets:
key = track_synset.keyParser(text_item, synset)
try:
key_vector = trained_model.word_vec(key)
context_vector.append(
key_vector
) # put all vector words in the sentence together
except KeyError:
pass
return numpy.average(context_vector, axis=0)
def test_cosine_similarity(self):
vec_manip = VectorManipulation()
a = [1, 0]
b = [-1, 0]
self.assertEqual(-1, vec_manip.cosine_similarity(a, b))
def __globalSimC(self, context_a, context_b):
vecmanip = VectorManipulation()
global_simc = vecmanip.cosine_similarity(context_a, context_b)
return global_simc
# python module absolute path
pydir_name = os.path.dirname(os.path.abspath(__file__))
ppydir_name = os.path.dirname(pydir_name)
# python path definition
sys.path.append(os.path.join(os.path.dirname(__file__), os.path.pardir))
# local-imports
from utilities.commandLine import CommandLineStats
from utilities.fileOperations import FileManipulation
from vecmanip.vectorOperations import VectorManipulation
if __name__ == '__main__':
params = CommandLineStats() # command line parameter validation
fio = FileManipulation()
stats_metrics = VectorManipulation()
gold_path = os.path.join(ppydir_name, params.gold_input)
ruby_path = os.path.join(ppydir_name, params.ruby_input)
ou_loc = os.path.join(ppydir_name, params.output_folder)
gold = fio.readFileLine(gold_path)
docs = fio.doclist_multifolder(ruby_path)
result = "Metric\tSpearman\tS-Rho\n"
for doc in docs:
ruby = fio.readFileLine(doc)
fname = doc.split(os.sep)
fname = fname[-1]
tmp_sp, tmp_rho = stats_metrics.spearmanCorrelation(gold, ruby)
result += fname[:-3] + '\t' + str(tmp_sp) + '\t' + str(tmp_rho) + '\n'