def getBestLambda(lambda_list_to_update_tmp):
global lambda_list_to_update
lambda_list_to_update = lambda_list_to_update_tmp
VtoE_model.createEntListTable_Spacy('dev')
EtoV_model.createEntListTable_Spacy('dev')
EtoV_model.createEntListTable_Stanford('dev')
VtoE_model.createEntListTable_Stanford('dev')
# CandidateSet.createCandidateSet(dev_list_EtoV,dev_list_VtoE,'dev')
ScoreTable.createScoreTable_TypeInSens(dev_list_EtoV,dev_list_VtoE,'dev')
ScoreTable.createScoreTable_TypeSens(dev_list_EtoV,dev_list_VtoE,'dev')
list_lambda = init_lambda()
best_lambda = list_lambda
best_res = init_result()
step = 0.1
while list_lambda != None:
print('List Lambda ', list_lambda)
cur_res = train_dev(list_lambda)
if (better_than(cur_res,best_res)):
best_lambda = dict((k,v) for k,v in list_lambda.items())
best_res = cur_res
list_lambda = update_list_lambda(list_lambda,step)
print('BestRes ' ,best_res)
print('BestLambda ', best_lambda)
return best_lambda
def createCandidateSet(EtoV_List, VtoE_List, mode):
'''
Create CandidateSet file
'''
EtoV_model.createEntListTable_Stanford(mode)
VtoE_model.createEntListTable_Stanford(mode)
EtoV_model.createEntListTable_Spacy(mode)
VtoE_model.createEntListTable_Spacy(mode)
print('Create Candidate Set')
if mode == 'dev':
candidate_set_file = candidate_set_dev_file
elif mode == 'test':
candidate_set_file = candidate_set_test_file
if os.path.isfile(candidate_set_file):
json_data = open(candidate_set_file).read()
Candidate_Set_Table = json.loads(json_data)
else:
Candidate_Set_Table = []
for i in range(len(EtoV_List)):
Candidate_Set_Table.append(
getCandidateSet(EtoV_List[i], VtoE_List[i], i))
with open(candidate_set_file, 'w', encoding='utf-8') as f:
json.dump(Candidate_Set_Table, f)
return Candidate_Set_Table
def main(lambda_list_to_update):
print(lambda_list_to_update)
list_lambda = training_TypeInSens.getBestLambda(lambda_list_to_update)
VtoE_model.createEntListTable_Spacy('test')
EtoV_model.createEntListTable_Spacy('test')
EtoV_model.createEntListTable_Stanford('test')
VtoE_model.createEntListTable_Stanford('test')
# list_lambda = config.getWeight()
# CandidateSet.createCandidateSet(test_list_EtoV,test_list_VtoE,'test')
# print("Create Candidate Set")
ScoreTable.createScoreTable_TypeInSens(test_list_EtoV,test_list_VtoE,'test')
ScoreTable.createScoreTable_TypeSens(test_list_EtoV,test_list_VtoE,'test')
print("Created Score Table")
print(list_lambda)
predict_set = getPredict.getFinalPredictNEPairList(test_list_EtoV, test_list_VtoE,list_lambda,'test',train_mode_InSens = True, train_mode_Sens=True)
true_set = TrueSet.getFileTrueSet(test_file_en,test_file_vn)
# print(predict_set[0])
# print(true_set[0])
# EvaluationRes = {'TP':,'TN':,}
# for i in range(len(predict_set)):
# print('=============')
# print('Predict', i , len(predict_set[i]))
# print(predict_set[i])
EvaluationRes_type_insen = evaluate_TypeInSens.getMetrics(predict_set,true_set,'test')
EvaluationRes_type_sen = evaluate_TypeSens.getMetrics(predict_set,true_set)
print('Type-insensitive ', EvaluationRes_type_insen)
print('Type-sensitive ', EvaluationRes_type_sen)
def getCandidateSet(EtoV_sent, VtoE_sent, sent_index):
'''
Get Candidate Set of 1 sentence pair
'''
EtoV_set = EtoV_model.getEntSetFromFile(EtoV_sent['Source'],
EtoV_sent['Target'], sent_index)
VtoE_set = VtoE_model.getEntSetFromFile(VtoE_sent['Source'],
VtoE_sent['Target'], sent_index)
V_Ent_List = []
E_Ent_List = []
for pair in EtoV_set:
V_Ent_List.append((pair[1], pair[2], pair[4]))
E_Ent_List.append((pair[0], pair[2], pair[3]))
for pair in VtoE_set:
E_Ent_List.append((pair[1], pair[2], pair[4]))
V_Ent_List.append((pair[0], pair[2], pair[3]))
res = []
for en_ent in E_Ent_List:
for vn_ent in V_Ent_List:
res.append((en_ent[0], vn_ent[0], en_ent[1], en_ent[2], vn_ent[2],
vn_ent[1]))
res = utilities.make_unique(res)
return res
def evaluateSentencePair(predict, true_set, mode):
'''
Compare predict set and true set of a sentence pair
'''
print('==================')
tp = 0
predict = utilities.make_unique(predict)
# print('evaluateSentencePair',predict)
predict = sorted(predict, key=lambda tmp: tmp[0])
# print('Predict Set', predict)
# print(predict)
# print('Predict Set ', predict)
# print('True Set ', true_set)
for i in range(len(true_set)):
true_en_begin = true_set[i][0][0]
# print('True_en_begin ',true_en_begin)
true_en_end = true_set[i][0][-1]
# print('True_en_end ',true_en_end)
sub_predict = []
for j in range(len(predict)):
if len(predict[j][0]) < 1:
continue
predict_en_begin = predict[j][0][0]
# print('PredictEnBegin ', predict_en_begin)
if predict_en_begin == true_en_begin:
sub_predict.append(predict[j])
# print('SubPredict ',sub_predict)
if (len(sub_predict)) > 0:
for pair in sub_predict:
# print(pair)
predict_en_end = pair[0][-1]
if predict_en_end == true_en_end:
if len(pair[1]) < 1:
continue
# print('Predict Pair',pair)
predict_vi_begin = pair[1][0]
predict_vi_end = pair[1][-1]
true_vi_begin = true_set[i][1][0]
true_vi_end = true_set[i][1][-1]
if true_vi_begin == predict_vi_begin and true_vi_end == predict_vi_end:
tp += 1
else:
candidateset = CandidateSet.getCandidateSetFromFile(
pair[-2], mode)
# print('CandidateSet', candidateset)
for k in range(len(candidateset)):
print('#####################')
print('Candidate', candidateset[k])
print(
'Score',
ScoreTable.getScoreforOneCandidate_TypeInSens(
pair[-2], k))
print('--------------')
print('True', true_set[i])
print('Predict', pair)
print(
'Score',
ScoreTable.getScoreforOneCandidate_TypeInSens(
pair[-2], pair[-1]))
print('Spacy Vi',
VtoE_model.getEntList_Spacy_FromFile(pair[-2]))
print('Spacy En',
EtoV_model.getEntList_Spacy_FromFile(pair[-2]))
print(
'Stanford En',
EtoV_model.getEntList_StanfordNER_FromFile(
pair[-2]))
print(
'Stanford Vi',
VtoE_model.getEntList_StanfordNER_FromFile(
pair[-2]))
# print('TP ', tp)
return tp