def get_threshold(We, words, Rel, rel, tm, relSize):
f = open('../commonsendata/Eval/conceptnet/new_omcs_dev1.txt', 'r')
lines = f.readlines()
Exp_S = []
r_list = []
t1_list = []
t2_list = []
for i in lines:
i = i.strip()
i = i.split('\t')
(r, t1, t2, score) = (i[0].strip(), i[1].strip(), i[2].strip(),
float(i[3]))
t1id = lookupwordID(We, words, t1)
t2id = lookupwordID(We, words, t2)
t1_list.append(t1id)
t2_list.append(t2id)
r_list.append(r)
t1batch = t1_list
t2batch = t2_list
x1, x1_mask, x1_length = tm.prepare_data(t1batch)
x2, x2_mask, x2_length = tm.prepare_data(t2batch)
v1 = tm.GetVector(x1, x1_mask, x1_length)
v2 = tm.GetVector(x2, x2_mask, x2_length)
for j in range(len(lines)):
v_r = Rel[rel[r_list[j].lower()], :].reshape((relSize))
input_vec = np.concatenate((v1[j], v_r, v2[j]), axis=0)
softmaxScore = tm.score_func(input_vec)
Exp_S.append(softmaxScore[0][0])
right = 0
wrong = 0
threshold = 0
accurancy = 0
binaryScore = []
Exp_S_sorted = sorted(Exp_S)
for j in xrange(len(Exp_S)):
temp_thr = Exp_S_sorted[j]
for j1 in xrange(int(len(Exp_S) / 2)):
if (Exp_S[j1] >= temp_thr):
right = right + 1
else:
wrong = wrong + 1
for j2 in xrange(int(len(Exp_S) / 2), int(len(Exp_S)), 1):
if (Exp_S[j2] < temp_thr):
right = right + 1
else:
wrong = wrong + 1
if ((right / (len(Exp_S))) > accurancy):
accurancy = (right / (len(Exp_S)))
threshold = temp_thr
right = 0
wrong = 0
print 'Dev1-Accurancy', accurancy
return threshold
def evaluate_conceptNet(We, words, Rel, rel, tm, memsize, relSize, fin):
threshold = get_threshold(We, words, Rel, rel, tm, memsize, relSize, fin)
f = open('../../commonsendata/Eval/conceptnet/new_omcs_dev2.txt', 'r')
lines = f.readlines()
Exp_S = []
r_list = []
t1_list = []
t2_list = []
for i in lines:
i = i.strip()
i = i.split('\t')
(r, t1, t2, score) = (i[0].strip(), i[1].strip(), i[2].strip(),
float(i[3]))
t1id = lookupwordID(We, words, t1)
t2id = lookupwordID(We, words, t2)
t1_list.append(t1id)
t2_list.append(t2id)
r_list.append(r)
t1batch = t1_list
t2batch = t2_list
delim = (lookupwordID(We, words, "#"))
batchTuple = [a + delim + b for a, b in zip(t1batch, t2batch)]
xx, xx_mask = tm.prepare_data(batchTuple)
vector = tm.GetVector(xx, xx_mask)
for j in range(len(lines)):
v_r = Rel[rel[r_list[j].lower()], :].reshape((1, relSize))
gvector = vector[j].reshape((1, memsize))
input_vec = np.concatenate((gvector, v_r), axis=1)
#input_vec = np.concatenate((gv1,v_r,gv2),axis = 1)
softmaxScore = tm.GetVectorNew(input_vec)
Exp_S.append(softmaxScore[0][0])
right = 0
wrong = 0
accurancy = 0
for j1 in xrange(int(len(Exp_S) / 2)):
if (Exp_S[j1] >= threshold):
right = right + 1
else:
wrong = wrong + 1
for j2 in xrange(int(len(Exp_S) / 2), int(len(Exp_S)), 1):
if (Exp_S[j2] < threshold):
right = right + 1
else:
wrong = wrong + 1
accurancy = (right / (len(Exp_S)))
print 'Dev2-Accurancy', accurancy
fin.write('Dev2-Accurancy' + str(accurancy) + "\n")
print 'Threshold', threshold
fin.write('Threshold' + str(threshold) + "\n")
return accurancy
def evaluate_conceptNet(words, We, rel, Rel, tm, relSize):
threshold, accurancy1 = get_threshold(
'../data/conceptnet/new_omcs_dev1.txt', We, words, rel, Rel, tm,
relSize)
f = open('../data/conceptnet/new_omcs_dev2.txt', 'r')
lines = f.readlines()
S = []
T = []
T1 = []
T2 = []
R = []
Exp_S = []
result = []
for i in lines:
i = i.strip()
i = i.split('\t')
(r, t1, t2, score) = (i[0].strip(), i[1].strip(), i[2].strip(),
float(i[3]))
temp1 = lookupwordID(We, words, t1)
temp2 = lookupwordID(We, words, t2)
T1.append(temp1)
T2.append(temp2)
tp = rel[r.lower()]
c = Rel[tp * relSize:tp * relSize + relSize, :]
R.append(c)
x1, x1_mask = tm.prepare_data(T1)
x2, x2_mask = tm.prepare_data(T2)
emb1 = tm.GetVector(x1, x1_mask)
emb2 = tm.GetVector(x2, x2_mask)
for j in range(len(R)):
gv1 = emb1[j]
gv2 = emb2[j]
v_r = R[j]
temp1 = np.dot(gv1, v_r)
exp_score = np.inner(temp1, gv2)
Exp_S.append(exp_score)
right = 0
wrong = 0
accurancy2 = 0
for j1 in xrange(int(len(Exp_S) / 2)):
if (Exp_S[j1] >= threshold):
right = right + 1
else:
wrong = wrong + 1
for j2 in xrange(int(len(Exp_S) / 2), int(len(Exp_S)), 1):
if (Exp_S[j2] <= threshold):
right = right + 1
else:
wrong = wrong + 1
accurancy2 = (right / (len(Exp_S)))
#print 'Dev2-Accurancy',accurancy
#print 'Threshold',threshold
return accurancy1, accurancy2, threshold
def evaluate_conceptNet(We, words, Rel, rel, tm, relSize):
threshold = get_threshold(We, words, Rel, rel, tm, relSize)
f = open('../commonsendata/Eval/conceptnet/new_omcs_dev2.txt', 'r')
lines = f.readlines()
Exp_S = []
r_list = []
t1_list = []
t2_list = []
tuple_words = []
tuple_vecs = []
for i in lines:
i = i.strip()
i = i.split('\t')
(r, t1, t2, score) = (i[0].strip(), i[1].strip(), i[2].strip(),
float(i[3]))
t1id = lookupwordID(We, words, t1)
t2id = lookupwordID(We, words, t2)
t1_list.append(t1id)
t2_list.append(t2id)
r_list.append(r)
tuple_words.append(i)
t1batch = t1_list[0:len(lines)]
t2batch = t2_list[0:len(lines)]
x1, x1_mask, x1_length = tm.prepare_data(t1batch)
x2, x2_mask, x2_length = tm.prepare_data(t2batch)
v1 = tm.GetVector(x1, x1_mask, x1_length)
v2 = tm.GetVector(x2, x2_mask, x2_length)
for j in range(len(lines)):
v_r = Rel[rel[r_list[j].lower()], :].reshape((relSize))
input_vec = np.concatenate((v1[j], v_r, v2[j]), axis=0)
softmaxScore = tm.score_func(input_vec)
tuple_vecs.append(tm.hidden_func(input_vec))
Exp_S.append(softmaxScore[0][0])
right = 0
wrong = 0
accurancy = 0
for j1 in xrange(int(len(Exp_S) / 2)):
if (Exp_S[j1] >= threshold):
right = right + 1
else:
wrong = wrong + 1
for j2 in xrange(int(len(Exp_S) / 2), int(len(Exp_S)), 1):
if (Exp_S[j2] < threshold):
right = right + 1
else:
wrong = wrong + 1
accurancy = (right / (len(Exp_S)))
print 'Dev2-Accurancy', accurancy
print 'Threshold', threshold
return accurancy, tuple_words, tuple_vecs
def evaluate_conceptNet(We, words, Rel, rel, tm, relSize):
threshold = get_threshold(We, words, Rel, rel, tm, relSize)
f = open('../commonsendata/Eval/conceptnet/new_omcs_dev2.txt', 'r')
lines = f.readlines()
Exp_S = []
r_list = []
t1_list = []
t2_list = []
for i in lines:
i = i.strip()
i = i.split('\t')
(r, t1, t2, score) = (i[0].strip(), i[1].strip(), i[2].strip(),
float(i[3]))
t1id = lookupwordID(We, words, t1)
t2id = lookupwordID(We, words, t2)
t1_list.append(t1id)
t2_list.append(t2id)
r_list.append(r)
t1batch = t1_list[0:len(lines)]
t2batch = t2_list[0:len(lines)]
x1, x1_mask, x1_length = tm.prepare_data(t1batch)
x2, x2_mask, x2_length = tm.prepare_data(t2batch)
v1 = tm.GetVector(x1, x1_mask, x1_length)
v2 = tm.GetVector(x2, x2_mask, x2_length)
for j in range(len(lines)):
v_r = Rel[rel[r_list[j].lower()] *
relSize:rel[r_list[j].lower()] * relSize + relSize, :]
temp1 = np.dot(v1[j], v_r)
exp_score = np.inner(temp1, v2[j])
Exp_S.append(exp_score)
right = 0
wrong = 0
accurancy = 0
for j1 in xrange(int(len(Exp_S) / 2)):
if (Exp_S[j1] >= threshold):
right = right + 1
else:
wrong = wrong + 1
for j2 in xrange(int(len(Exp_S) / 2), int(len(Exp_S)), 1):
if (Exp_S[j2] < threshold):
right = right + 1
else:
wrong = wrong + 1
accurancy = (right / (len(Exp_S)))
print 'Dev2-Accurancy', accurancy
print 'Threshold', threshold
return accurancy
def get_accu(We, words, Rel, rel, tm, relSize, threshold, filename):
f = open(filename, 'r')
lines = f.readlines()
lines.append('ReceivesAction\thockey\tplay on ice\t1')
lines.append('AtLocation\trestroom\trest area\t1')
Exp_S = []
r_list = []
t1_list = []
t2_list = []
for i in lines:
i = i.strip()
i = i.split('\t')
(r, t1, t2, score) = (i[0].strip(), i[1].strip(), i[2].strip(),
float(i[3]))
t1id = lookupwordID(We, words, t1)
t2id = lookupwordID(We, words, t2)
t1_list.append(t1id)
t2_list.append(t2id)
r_list.append(r)
t1batch = t1_list[0:len(lines)]
t2batch = t2_list[0:len(lines)]
x1, x1_mask, x1_length = tm.prepare_data(t1batch)
x2, x2_mask, x2_length = tm.prepare_data(t2batch)
v1 = tm.GetVector(x1, x1_mask, x1_length)
v2 = tm.GetVector(x2, x2_mask, x2_length)
for j in range(len(lines)):
v_r = Rel[rel[r_list[j].lower()], :].reshape((relSize))
input_vec = np.concatenate((v1[j], v_r, v2[j]), axis=0)
softmaxScore = tm.score_func(input_vec)
# if j == len(lines)-1:
# print lines[j],softmaxScore
Exp_S.append(softmaxScore[0][0])
right = 0
wrong = 0
accurancy = 0
for j1 in xrange(int(len(Exp_S) / 2)):
if (Exp_S[j1] >= threshold):
right = right + 1
else:
wrong = wrong + 1
for j2 in xrange(int(len(Exp_S) / 2), int(len(Exp_S)), 1):
if (Exp_S[j2] < threshold):
right = right + 1
else:
wrong = wrong + 1
accurancy = (right / (len(Exp_S)))
return accurancy
def get_accu(We, words, Rel, rel, tm, relSize, threshold, filename):
f = open(filename, 'r')
lines = f.readlines()
Exp_S = []
r_list = []
t1_list = []
t2_list = []
for i in lines:
i = i.strip()
i = i.split('\t')
(r, t1, t2, score) = (i[0].strip(), i[1].strip(), i[2].strip(),
float(i[3]))
t1id = lookupwordID(We, words, t1)
t2id = lookupwordID(We, words, t2)
t1_list.append(t1id)
t2_list.append(t2id)
r_list.append(r)
t1batch = t1_list[0:len(lines)]
t2batch = t2_list[0:len(lines)]
x1, x1_mask, x1_length = tm.prepare_data(t1batch)
x2, x2_mask, x2_length = tm.prepare_data(t2batch)
v1 = tm.GetVector(x1, x1_mask, x1_length)
v2 = tm.GetVector(x2, x2_mask, x2_length)
for j in range(len(lines)):
v_r = Rel[rel[r_list[j].lower()] *
relSize:rel[r_list[j].lower()] * relSize + relSize, :]
temp1 = np.dot(v1[j], v_r)
exp_score = np.inner(temp1, v2[j])
Exp_S.append(exp_score)
right = 0
wrong = 0
accurancy = 0
for j1 in xrange(int(len(Exp_S) / 2)):
if (Exp_S[j1] >= threshold):
right = right + 1
else:
wrong = wrong + 1
for j2 in xrange(int(len(Exp_S) / 2), int(len(Exp_S)), 1):
if (Exp_S[j2] < threshold):
right = right + 1
else:
wrong = wrong + 1
accurancy = (right / (len(Exp_S)))
return accurancy
def getpairs2(self, batch, params):
embed_size = self.memsize
Rel = self.getRel()
we = self.getWe()
# Rel0 = np.reshape(Rel,(-1,relsize))
newd = [
convertToIndex(i, self.words, we, self.rel, Rel) for i in batch
]
g1 = []
g2 = []
R = []
#print newd
length = len(batch)
for idx, e in enumerate(newd):
(r, t1, t2, s) = e
g1.append(t1)
g2.append(t2)
R.append(r)
#batch is list of tuples
p11 = []
p22 = []
p3 = []
if (params.type == 'MAX'):
for i in range(length):
#print 'i: ',i
id0 = R[i]
wpick = ['', '', '']
while (wpick[0] == ''):
index = random.randint(0, len(g1) - 1)
if (index != i):
wpick[0] = g1[index]
p11.append(wpick[0])
while (wpick[1] == ''):
index = random.randint(0, len(g2) - 1)
if (index != i):
wpick[1] = g2[index]
p22.append(wpick[1])
while (wpick[2] == ''):
index = random.randint(0, len(R) - 1)
if (index != i):
wpick[2] = R[index]
p3.append(wpick[2])
delim = (lookupwordID(we, self.words, "#"))
pT = [a + delim + b for a, b in zip(g1, g2)]
pTuple, pTupleMask = self.prepare_data(pT)
neT1 = [a + delim + b for a, b in zip(p11, g2)]
neTuple1, neTuple1Mask = self.prepare_data(neT1)
neT2 = [a + delim + b for a, b in zip(g1, p22)]
neTuple2, neTuple2Mask = self.prepare_data(neT2)
return (R, p3, pTuple, pTupleMask, neTuple1, neTuple1Mask, neTuple2,
neTuple2Mask)
def prepare_aedata(self, list_of_seqs, contextsize, words):
lengths = [len(s) for s in list_of_seqs]
sumlength = sum(lengths)
n_samples = len(list_of_seqs)
D = np.zeros((sumlength, 10)).astype('int32')
x = np.zeros((sumlength, 2 * contextsize + 1)).astype('int32')
index = 0
for i in range(n_samples):
seq_id = lookupwordID(words, list_of_seqs[i])
new_seq = [0] * contextsize + seq_id + [1] * contextsize
for j in range(lengths[i]):
x[index, :] = new_seq[j:j + 2 * contextsize + 1]
word_j = list_of_seqs[i][j]
if len(word_j) == 0:
print 'error'
print list_of_seqs[i]
sys.exit()
punc_flag = 1
a = 0
for s in word_j:
if s in string.punctuation:
a = a + 1
if a == len(word_j):
punc_flag = 0
if word_j == '<@MENTION>':
D[index, 0] = 1
elif (word_j[0] == '#') and (len(word_j) != 1):
D[index, 1] = 1
elif word_j == 'rt':
D[index, 2] = 1
elif 'URL' in word_j:
D[index, 3] = 1
elif word_j.replace('.', '', 1).isdigit():
D[index, 4] = 1
# check whether it is punc
elif '$' in word_j:
D[index, 5] = 1
elif word_j == ':':
D[index, 7] = 1
elif word_j == '...':
D[index, 8] = 1
elif (len(word_j) == 1) and (word_j[0] in string.punctuation):
D[index, 9] = 1
elif punc_flag == 0:
D[index, 6] = 1
index = index + 1
#print len(labels)
return x, n_samples, D
def get_threshold(evafile, We, words, rel, Rel, tm, relSize):
f1 = open(evafile, 'r')
lines = f1.readlines()
S = []
T = []
T1 = []
T2 = []
R = []
Exp_S = []
result = []
for i in lines:
i = i.strip()
i = i.split('\t')
(r, t1, t2, score) = (i[0].strip(), i[1].strip(), i[2].strip(),
float(i[3]))
temp1 = lookupwordID(We, words, t1)
temp2 = lookupwordID(We, words, t2)
T1.append(temp1)
T2.append(temp2)
tp = rel[r.lower()]
c = Rel[tp * relSize:tp * relSize + relSize, :]
R.append(c)
x1, x1_mask = tm.prepare_data(T1)
x2, x2_mask = tm.prepare_data(T2)
emb1 = tm.GetVector(x1, x1_mask)
emb2 = tm.GetVector(x2, x2_mask)
for j in range(len(R)):
gv1 = emb1[j]
gv2 = emb2[j]
v_r = R[j]
temp1 = np.dot(gv1, v_r)
exp_score = np.inner(temp1, gv2)
Exp_S.append(exp_score)
right = 0
wrong = 0
threshold = 0
accurancy = 0
binaryScore = []
Exp_S_sorted = sorted(Exp_S)
for j in xrange(len(Exp_S)):
temp_thr = Exp_S_sorted[j]
for j1 in xrange(int(len(Exp_S) / 2)):
if (Exp_S[j1] >= temp_thr):
right = right + 1
else:
wrong = wrong + 1
for j2 in xrange(int(len(Exp_S) / 2), int(len(Exp_S)), 1):
if (Exp_S[j2] <= temp_thr):
right = right + 1
else:
wrong = wrong + 1
if ((right / (len(Exp_S))) > accurancy):
accurancy = (1.0 * right / (len(Exp_S)))
threshold = temp_thr
right = 0
wrong = 0
#print 'Dev1-Accurancy',accurancy
return threshold, accurancy
def evaCOPA(evafile, words, We, rel, Rel, evaType, tm, relSize):
f = open(evafile, 'r')
lines = f.readlines()
trueAns = []
q = []
alter1 = []
alter2 = []
causality = []
same = 0
diff = 0
totalScore1 = []
totalScore2 = []
for i in xrange(4, len(lines) - 1, 6):
singleAns = lines[i][lines[i].find('alternative=') +
13:lines[i].find('>') - 1]
trueAns.append(singleAns)
if (lines[i].find('effect') != -1):
causality.append(1)
else:
causality.append(0)
singleq = lines[i + 1][lines[i + 1].find('<p>') +
3:lines[i + 1].find('</p>') - 1]
temp0 = lookupwordID(We, words, singleq)
q.append(temp0)
singleAlter1 = lines[i + 2][lines[i + 2].find('<a1>') +
4:lines[i + 2].find('</a1>') - 1]
temp1 = lookupwordID(We, words, singleAlter1)
alter1.append(temp1)
singleAlter2 = lines[i + 3][lines[i + 3].find('<a2>') +
4:lines[i + 3].find('</a2>') - 1]
temp2 = lookupwordID(We, words, singleAlter2)
alter2.append(temp2)
idx = 0
while idx < len(q):
qq = q[idx:idx + 100 if idx + 100 < len(q) else len(q)]
alter1q = alter1[idx:idx + 100 if idx + 100 < len(q) else len(q)]
alter2q = alter2[idx:idx + 100 if idx + 100 < len(q) else len(q)]
causalityq = causality[idx:idx + 100 if idx + 100 < len(q) else len(q)]
x0, x0_mask = tm.prepare_data(qq)
x1, x1_mask = tm.prepare_data(alter1q)
x2, x2_mask = tm.prepare_data(alter2q)
emb0 = tm.GetVector(x0, x0_mask)
emb1 = tm.GetVector(x1, x1_mask)
emb2 = tm.GetVector(x2, x2_mask)
for j in range(100):
if (causalityq[j] == 0):
scores1 = score(emb1[j], emb0[j], words, We, rel, Rel, relSize)
scores2 = score(emb2[j], emb0[j], words, We, rel, Rel, relSize)
if (evaType.lower() == 'max'):
score1 = scores1[0]
score2 = scores2[0]
if (evaType.lower() == 'sum'):
score1 = scores1[1]
score2 = scores2[1]
if (evaType.lower() == 'cause'):
score1 = scores1[2]
score2 = scores2[2]
totalScore1.append(scores1)
totalScore2.append(scores2)
else:
scores1 = score(emb0[j], emb1[j], words, We, rel, Rel, relSize)
scores2 = score(emb0[j], emb2[j], words, We, rel, Rel, relSize)
if (evaType.lower() == 'max'):
score1 = scores1[0]
score2 = scores2[0]
if (evaType.lower() == 'sum'):
score1 = scores1[1]
score2 = scores2[1]
if (evaType.lower() == 'cause'):
score2 = scores1[2]
score2 = scores2[2]
totalScore1.append(scores1)
totalScore2.append(scores2)
if (score1 > score2):
ans = 1
else:
ans = 2
if (ans == int(trueAns[idx])):
same = same + 1
else:
diff = diff + 1
idx = idx + 100
# print 'totalScore1',len(totalScore1)
# print 'totalScore2',len(totalScore2)
# print 'trueAns',len(trueAns)
#print same, diff
return same / (same + diff), totalScore1, totalScore2, trueAns
def get_threshold(We, words, Rel, rel, tm, relSize, dev1_file):
f = open(dev1_file, 'r')
lines = f.readlines()
Exp_S = []
r_list = []
t1_list = []
t2_list = []
for i in lines:
i = i.strip()
i = i.split('\t')
(r, t1, t2, score) = (i[0].strip(), i[1].strip(), i[2].strip(),
float(i[3]))
t1id = lookupwordID(We, words, t1)
t2id = lookupwordID(We, words, t2)
t1_list.append(t1id)
t2_list.append(t2id)
r_list.append(r)
t1batch = t1_list
t2batch = t2_list
x1, x1_mask, x1_length = tm.prepare_data(t1batch)
x2, x2_mask, x2_length = tm.prepare_data(t2batch)
v1 = tm.GetVector(x1, x1_mask, x1_length)
v2 = tm.GetVector(x2, x2_mask, x2_length)
for j in range(len(lines)):
v_r = Rel[rel[r_list[j].lower()] *
relSize:rel[r_list[j].lower()] * relSize + relSize, :]
temp1 = np.dot(v1[j], v_r)
exp_score = np.inner(temp1, v2[j])
Exp_S.append(exp_score)
right = 0
wrong = 0
threshold = 0
accurancy = 0
binaryScore = []
Exp_S_sorted = sorted(Exp_S)
for j in xrange(len(Exp_S)):
temp_thr = Exp_S_sorted[j]
for j1 in xrange(int(len(Exp_S) / 2)):
if (Exp_S[j1] >= temp_thr):
right = right + 1
else:
wrong = wrong + 1
for j2 in xrange(int(len(Exp_S) / 2), int(len(Exp_S)), 1):
if (Exp_S[j2] < temp_thr):
right = right + 1
else:
wrong = wrong + 1
if ((right / (len(Exp_S))) > accurancy):
accurancy = (right / (len(Exp_S)))
threshold = temp_thr
right = 0
wrong = 0
print 'Dev1-Accurancy', accurancy
return threshold
def get_threshold(We, words, Rel, rel, tm, memsize, relSize, fin):
f = open('../../commonsendata/Eval/conceptnet/new_omcs_dev1.txt', 'r')
lines = f.readlines()
Exp_S = []
r_list = []
t1_list = []
t2_list = []
for i in lines:
i = i.strip()
i = i.split('\t')
(r, t1, t2, score) = (i[0].strip(), i[1].strip(), i[2].strip(),
float(i[3]))
t1id = lookupwordID(We, words, t1)
t2id = lookupwordID(We, words, t2)
t1_list.append(t1id)
t2_list.append(t2id)
r_list.append(r)
t1batch = t1_list[0:len(lines)]
t2batch = t2_list[0:len(lines)]
print 't1batch: ', len(t1batch)
delim = (lookupwordID(We, words, "#"))
batchTuple = [a + delim + b for a, b in zip(t1batch, t2batch)]
xx, xx_mask = tm.prepare_data(batchTuple)
vector = tm.GetVector(xx, xx_mask)
for j in range(len(lines)):
v_r = Rel[rel[r_list[j].lower()], :].reshape((1, relSize))
vectorg = vector[j].reshape((1, memsize))
input_vec = np.concatenate((vectorg, v_r), axis=1)
softmaxScore = tm.GetVectorNew(input_vec)
Exp_S.append(softmaxScore[0][0])
right = 0
wrong = 0
threshold = 0
accurancy = 0
binaryScore = []
Exp_S_sorted = sorted(Exp_S)
for j in xrange(len(Exp_S)):
temp_thr = Exp_S_sorted[j]
for j1 in xrange(int(len(Exp_S) / 2)):
if (Exp_S[j1] >= temp_thr):
right = right + 1
else:
wrong = wrong + 1
for j2 in xrange(int(len(Exp_S) / 2), int(len(Exp_S)), 1):
if (Exp_S[j2] < temp_thr):
right = right + 1
else:
wrong = wrong + 1
if ((right / (len(Exp_S))) > accurancy):
accurancy = (right / (len(Exp_S)))
threshold = temp_thr
right = 0
wrong = 0
print 'Dev1-Accurancy', accurancy
fin.write('Dev1-Accurancy' + str(accurancy) + "\n")
return threshold
