def test_model_global(nn_model,
testdata,
chardata,
pos_data,
index2word,
resultfile='',
batch_size=50):
index2word[0] = ''
index2word_BIOES = {0: '', 1: 'B', 2: 'I', 3: 'O', 4: 'E', 5: 'S'}
index2word_Type = {0: '', 1: 'O', 2: 'LOC', 3: 'ORG', 4: 'PER', 5: 'MISC'}
testx = np.asarray(testdata[0], dtype="int32")
testy = np.asarray(testdata[1], dtype="int32")
testy_BIOES = np.asarray(testdata[3], dtype="int32")
testy_Type = np.asarray(testdata[4], dtype="int32")
poslabel_test = np.asarray(pos_data, dtype="int32")
testchar = np.asarray(chardata, dtype="int32")
testresult = []
testresult2 = []
testresult3 = []
predictions = nn_model.predict([testx, testchar])
isfinall = False
for si in range(0, len(predictions[0])):
ptag = []
for iw, word in enumerate(predictions[0][si]):
next_index = np.argmax(word)
next_token = index2word_BIOES[next_index]
next_index2 = np.argmax(predictions[1][si][iw])
next_token2 = index2word_Type[next_index2]
if next_token == 'O' or next_token == '':
ptag.append(next_token)
else:
ptag.append(next_token + '-' + next_token2)
ttag = []
for word in testy[si]:
next_index = np.argmax(word)
next_token = index2word[next_index]
ttag.append(next_token)
result3 = []
result3.append(ptag)
result3.append(ttag)
testresult3.append(result3)
P, R, F, PR_count, P_count, TR_count = evaluation_NER(
testresult3, resultfile=resultfile)
print('global---Type>>>>>>>>>>', P, R, F)
return P, R, F, PR_count, P_count, TR_count
def test_model(nn_model,
inputs_test_x,
test_y,
index2word,
resultfile='',
batch_size=10):
index2word[0] = ''
predictions = nn_model.predict(inputs_test_x)
testresult = []
for si in range(0, len(predictions)):
sent = predictions[si]
# print('predictions',sent)
ptag = []
for word in sent:
next_index = np.argmax(word)
# if next_index != 0:
next_token = index2word[next_index]
ptag.append(next_token)
# print('next_token--ptag--',str(ptag))
senty = test_y[0][si]
ttag = []
# flag =0
for word in senty:
next_index = np.argmax(word)
next_token = index2word[next_index]
# if word > 0:
# if flag == 0:
# flag = 1
# count+=1
ttag.append(next_token)
# print(si, 'next_token--ttag--', str(ttag))
result = []
result.append(ptag)
result.append(ttag)
testresult.append(result)
# print(result.shape)
# print('count-----------',count)
# pickle.dump(testresult, open(resultfile, 'w'))
# P, R, F = evaluavtion_triple(testresult)
P, R, F, PR_count, P_count, TR_count = evaluation_NER(testresult)
# evaluation_NER2(testresult)
# print (P, R, F)
# evaluation_NER_error(testresult)
return P, R, F, PR_count, P_count, TR_count
def test_model(nn_model,
testdata,
chardata,
pos_data,
index2word,
resultfile='',
batch_size=50):
index2word[0] = ''
testx = np.asarray(testdata[0], dtype="int32")
testy = np.asarray(testdata[1], dtype="int32")
testpos = np.asarray(pos_data, dtype="int32")
testchar = np.asarray(chardata, dtype="int32")
testresult = []
predictions = nn_model.predict([testx, testchar])
for si in range(0, len(predictions)):
sent = predictions[si]
ptag = []
for word in sent:
next_index = np.argmax(word)
next_token = index2word[next_index]
ptag.append(next_token)
# print('next_token--ptag--',str(ptag))
senty = testy[si]
ttag = []
for word in senty:
next_index = np.argmax(word)
next_token = index2word[next_index]
ttag.append(next_token)
result = []
result.append(ptag)
result.append(ttag)
testresult.append(result)
P, R, F, PR_count, P_count, TR_count = evaluation_NER(
testresult, resultfile)
return P, R, F, PR_count, P_count, TR_count
def test_model(nn_model,
inputs_test_x,
test_y,
index2word,
resultfile='',
batch_size=10,
testfile=''):
index2word[0] = ''
predictions = nn_model.predict(inputs_test_x)
testresult = []
for si in range(0, len(predictions)):
sent = predictions[si]
# print('predictions',sent)
ptag = []
for word in sent:
next_index = np.argmax(word)
# if next_index == 0:
# break
next_token = index2word[next_index]
ptag.append(next_token)
senty = test_y[0][si]
ttag = []
for word in senty:
next_index = np.argmax(word)
if next_index == 0:
break
next_token = index2word[next_index]
ttag.append(next_token)
result = []
result.append(ptag)
result.append(ttag)
testresult.append(result)
P, R, F, PR_count, P_count, TR_count = evaluation_NER(testresult)
return P, R, F, PR_count, P_count, TR_count
def test_model(nn_model, index2word, batch_size=10):
# index2word[0] = ''
test_D = data_generator_4test(test,
test_posi,
batch_size=batch_size,
maxlen=max_s)
predictions = nn_model.predict_generator(generator=test_D.__iter__(),
steps=len(test_D),
verbose=1)
print(len(test), len(predictions))
testresult = []
for si in range(0, len(predictions)):
sent = predictions[si]
ptag = []
senty = test_label[si]
ttag = []
for wi, word in enumerate(senty):
next_index = np.argmax(word)
next_token = index2word[next_index]
ttag.append(next_token)
next_index = np.argmax(sent[wi])
next_token = index2word[next_index]
ptag.append(next_token)
result = []
result.append(ptag)
result.append(ttag)
testresult.append(result)
P, R, F, PR_count, P_count, TR_count = evaluation_NER(testresult)
return P, R, F, PR_count, P_count, TR_count
def test_model_segment(nn_model, testdata, chartest, index2tag):
index2tag[0] = ''
testx = np.asarray(testdata[0], dtype="int32")
testy_BIOES = np.asarray(testdata[1], dtype="int32")
testchar = np.asarray(chartest, dtype="int32")
predictions = nn_model.predict([testx, testchar])
testresult_1Step = []
testresult2 = []
for si in range(0, len(predictions)):
ptag_BIOES = []
ptag_1Step = []
for word in predictions[si]:
next_index = np.argmax(word)
next_token = index2tag[next_index]
ptag_BIOES.append(next_token)
if next_token != '':
ptag_1Step.append(next_token)
ttag_BIOES = []
for word in testy_BIOES[si]:
next_index = np.argmax(word)
next_token = index2tag[next_index]
ttag_BIOES.append(next_token)
testresult2.append([ptag_BIOES, ttag_BIOES])
testresult_1Step.append(ptag_1Step)
P, R, F, PR_count, P_count, TR_count = evaluation_NER(testresult2,
resultfile='')
print('NER test results >>>>>>>>>>', P, R, F, PR_count, P_count, TR_count)
return testresult_1Step
def test_model_divide(nn_model,
testdata,
chardata,
pos_data,
index2word,
resultfile='',
batch_size=50):
index2word[0] = ''
index2word_BIOES = {0: '', 1: 'B', 2: 'I', 3: 'O', 4: 'E', 5: 'S'}
index2word_Type = {0: '', 1: 'O', 2: 'LOC', 3: 'ORG', 4: 'PER', 5: 'MISC'}
testx = np.asarray(testdata[0], dtype="int32")
testy = np.asarray(testdata[1], dtype="int32")
testy_BIOES = np.asarray(testdata[3], dtype="int32")
testy_Type = np.asarray(testdata[4], dtype="int32")
poslabel_test = np.asarray(pos_data, dtype="int32")
testchar = np.asarray(chardata, dtype="int32")
testresult = []
testresult2 = []
testresult3 = []
predictions = nn_model.predict([testx, testchar])
isfinall = False
for si in range(0, len(predictions[0])):
ptag_BIOES = []
for word in predictions[0][si]:
next_index = np.argmax(word)
next_token = index2word_BIOES[next_index]
ptag_BIOES.append(next_token)
# print('next_token--ptag--',str(ptag))
ptag_Type = []
for word in predictions[1][si]:
next_index = np.argmax(word)
if len(word) == 6:
next_token = index2word_Type[next_index]
else:
isfinall = True
next_token = index2word[next_index]
ptag_Type.append(next_token)
# print('next_token--ptag--',str(ptag))
ttag_BIOES = []
for word in testy_BIOES[si]:
next_index = np.argmax(word)
next_token = index2word_BIOES[next_index]
ttag_BIOES.append(next_token)
ttag_Type = []
if isfinall:
for word in testy[si]:
next_index = np.argmax(word)
next_token = index2word[next_index]
ttag_Type.append(next_token)
else:
for word in testy_Type[si]:
next_index = np.argmax(word)
next_token = index2word_Type[next_index]
ttag_Type.append(next_token)
result2 = []
result2.append(ptag_BIOES)
result2.append(ttag_BIOES)
testresult2.append(result2)
result3 = []
result3.append(ptag_Type)
result3.append(ttag_Type)
testresult3.append(result3)
P, R, F, PR_count, P_count, TR_count = evaluation_NER_BIOES(
testresult2, resultfile=resultfile + '.BIORS.txt')
print('divide---BIOES>>>>>>>>>>', P, R, F)
if isfinall:
P, R, F, PR_count, P_count, TR_count = evaluation_NER(
testresult3, resultfile=resultfile)
else:
P, R, F, PR_count, P_count, TR_count = evaluation_NER_Type(
testresult3, resultfile=resultfile + '.Type.txt')
print('divide---Type>>>>>>>>>>', P, R, F)
return P, R, F, PR_count, P_count, TR_count
