def main(_):
loadData = False
useOntology = True # Ontology
runCABASC = False
runLCRROT = False
runLCRROTINVERSE = False
runLCRROTALT = False #Olaf model
runSVM = False
runLCRROTALT_v4 = False # Maria Model
weightanalysis = False
#determine if backupmethod is used
if runCABASC or runLCRROT or runLCRROTALT or runLCRROTINVERSE or runSVM:
backup = True
else:
backup = False
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector = loadDataAndEmbeddings(
FLAGS, loadData)
print(test_size)
remaining_size = 250
accuracyOnt = 0.87
if useOntology == True:
print('Starting Ontology Reasoner')
Ontology = OntReasoner()
#out of sample accuracy
accuracyOnt, remaining_size = Ontology.run(backup, FLAGS.test_path,
runSVM)
#in sample accuracy
Ontology = OntReasoner()
accuracyInSampleOnt, remaining_size = Ontology.run(
backup, FLAGS.train_path, runSVM)
if runSVM == True:
test = FLAGS.remaining_svm_test_path
else:
test = FLAGS.remaining_test_path
print('train acc = {:.4f}, test acc={:.4f}, remaining size={}'.format(
accuracyInSampleOnt, accuracyOnt, remaining_size))
else:
if runSVM == True:
test = FLAGS.test_svm_path
else:
test = FLAGS.test_path
if runLCRROTALT_v4 == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt_hierarchical_v4.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
# LCR-Rot-hop model
if runLCRROTALT == True:
tf.reset_default_graph()
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
print([_, pred2, fw2, bw2, tl2, tr2])
print('Finished program succesfully')
def main(_):
loadData = True # only for non-contextualised word embeddings.
# Use prepareBERT for BERT (and BERT_Large) and prepareELMo for ELMo
useOntology = False # When run together with runLCRROTALT, the two-step method is used
runLCRROTALT = True
runSVM = False
runCABASC = False
runLCRROT = False
runLCRROTINVERSE = False
weightanalysis = False
#determine if backupmethod is used
if runCABASC or runLCRROT or runLCRROTALT or runLCRROTINVERSE or runSVM:
backup = True
else:
backup = False
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector = loadDataAndEmbeddings(
FLAGS, loadData)
print(test_size)
remaining_size = 250
accuracyOnt = 0.87
if useOntology == True:
print('Starting Ontology Reasoner')
Ontology = OntReasoner()
#out of sample accuracy
accuracyOnt, remaining_size = Ontology.run(backup, FLAGS.test_path_ont,
runSVM)
#in sample accuracy
Ontology = OntReasoner()
accuracyInSampleOnt, remaining_size = Ontology.run(
backup, FLAGS.train_path_ont, runSVM)
if runSVM == True:
test = FLAGS.remaining_svm_test_path
else:
test = FLAGS.remaining_test_path
print(test[0])
print('train acc = {:.4f}, test acc={:.4f}, remaining size={}'.format(
accuracyOnt, accuracyOnt, remaining_size))
else:
if runSVM == True:
test = FLAGS.test_svm_path
else:
test = FLAGS.test_path
# LCR-Rot-hop model
if runLCRROTALT == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
def main(_):
loadData = False
useOntology = True
runCABASC = False
runLCRROT = False
runLCRROTINVERSE = False
runLCRROTALT = True
runSVM = False
#determine if backupmethod is used
if runCABASC or runLCRROT or runLCRROTALT or runLCRROTINVERSE or runSVM:
backup = True
else:
backup = False
BASE_train = "C:/Users/Maria/Desktop/data/programGeneratedData/crossValidation"+str(FLAGS.year)+'/cross_train_'
BASE_val = "C:/Users/Maria/Desktop/data/programGeneratedData/crossValidation"+str(FLAGS.year)+'/cross_val_'
BASE_svm_train = "C:/Users/Maria/Desktop/data/programGeneratedData/crossValidation"+str(FLAGS.year)+'/svm/cross_train_svm_'
BASE_svm_val = "C:/Users/Maria/Desktop/data/programGeneratedData/crossValidation"+str(FLAGS.year)+'/svm/cross_val_svm_'
REMAIN_val = "C:/Users/Maria/Desktop/data/programGeneratedData/crossValidation"+str(FLAGS.year)+'/cross_val_remainder_'
REMAIN_svm_val = "C:/Users/Maria/Desktop/data/programGeneratedData/crossValidation"+str(FLAGS.year)+'/svm/cross_val_remainder_'
# Number of k-fold cross validations
split_size = 10
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector = loadCrossValidation(FLAGS, split_size, loadData)
remaining_size = 248
accuracyOnt = 0.87
if useOntology == True:
print('Starting Ontology Reasoner')
acc = []
remaining_size_vec = []
#k-fold cross validation
for i in range(split_size):
Ontology = OntReasoner()
accuracyOnt, remaining_size = Ontology.run(backup,BASE_val+str(i)+'.txt', runSVM, True, i)
acc.append(accuracyOnt)
remaining_size_vec.append(remaining_size)
with open("C:/Users/Maria/Desktop/data/programGeneratedData/crossValidation"+str(FLAGS.year)+'/cross_results_"+str(FLAGS.year)+"/ONTOLOGY_"+str(FLAGS.year)+'.txt', 'w') as result:
print(str(split_size)+'-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print(acc)
result.write('size:' + str(test_size))
result.write('accuracy: '+ str(acc)+'\n')
result.write('remaining size: '+ str(remaining_size_vec)+'\n')
result.write('Accuracy: {}, St Dev:{} \n'.format(np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
if runSVM == True:
test = REMAIN_svm_val
else:
test = REMAIN_val
def main(_):
loadData = True # only for non-contextualised word embeddings.
augment_data = True # Load data must be true to augment
# Use prepareBERT for BERT (and BERT_Large) and prepareELMo for ELMo
useOntology = False # When run together with runLCRROTALT, the two-step method is used
runLCRROTALT = True
# determine if backupmethod is used
if runLCRROTALT:
backup = True
else:
backup = False
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector, ct = loadDataAndEmbeddings(
FLAGS, loadData, augment_data)
remaining_size = 250
accuracyOnt = 0.87 if FLAGS.year == 2016 else 0.8277
if useOntology == True:
print('Starting Ontology Reasoner')
# in sample accuracy
Ontology = OntReasoner()
accuracyOnt, remaining_size = Ontology.run(backup, FLAGS.test_path_ont,
runSVM)
# out of sample accuracy
# Ontology = OntReasoner()
# accuracyInSampleOnt, remainingInSample_size = Ontology.run(backup,FLAGS.train_path_ont, runSVM)
test = FLAGS.remaining_test_path
print(test[0])
print('train acc = {:.4f}, test acc={:.4f}, remaining size={}'.format(
accuracyOnt, accuracyOnt, remaining_size))
else:
test = FLAGS.remaining_test_path
# LCR-Rot-hop model
if runLCRROTALT == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size,
augment_data, FLAGS.augmentation_file_path, ct)
tf.reset_default_graph()
def main(_):
loadData = False
useOntology = False
runCABASC = False
runLCRROT = False
runLCRROTINVERSE = False
runLCRROTALT = False
runSVM = False
runLCRModelAlt_hierarchical_v4 = True
runAdversarial = True
#determine if backupmethod is used
if runCABASC or runLCRROT or runLCRROTALT or runLCRROTINVERSE or runSVM:
backup = True
else:
backup = False
BASE_train = FLAGS.hardcoded_path + "/data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/cross_train_'
BASE_val = FLAGS.hardcoded_path + "/data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/cross_val_'
BASE_svm_train = FLAGS.hardcoded_path + "/data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/svm/cross_train_svm_'
BASE_svm_val = FLAGS.hardcoded_path + "/data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/svm/cross_val_svm_'
REMAIN_val = FLAGS.hardcoded_path + "/data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/cross_val_remainder_'
REMAIN_svm_val = FLAGS.hardcoded_path + "/data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/svm/cross_val_remainder_'
# Number of k-fold cross validations
split_size = 10
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector = loadCrossValidation(
FLAGS, split_size, loadData)
remaining_size = 248
accuracyOnt = 0.87
if useOntology == True:
print('Starting Ontology Reasoner')
acc = []
remaining_size_vec = []
#k-fold cross validation
for i in range(split_size):
Ontology = OntReasoner()
accuracyOnt, remaining_size = Ontology.run(
backup, BASE_val + str(i) + '.txt', runSVM, True, i)
acc.append(accuracyOnt)
remaining_size_vec.append(remaining_size)
with open(
FLAGS.hardcoded_path +
"/data/programGeneratedData/crossValidation" +
str(FLAGS.year) +
'/cross_results_"+str(FLAGS.year)+"/ONTOLOGY_"+str(FLAGS.year)'
+ '.txt', 'w') as result:
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
print(acc)
result.write('size:' + str(test_size))
result.write('accuracy: ' + str(acc) + '\n')
result.write('remaining size: ' + str(remaining_size_vec) + '\n')
result.write('Accuracy: {}, St Dev:{} \n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
if runSVM == True:
test = REMAIN_svm_val
else:
test = REMAIN_val
else:
if runSVM == True:
test = BASE_svm_val
else:
test = BASE_val
#test = REMAIN_val
if runLCRROT == True:
acc = []
#k-fold cross validation
for i in [8]:
acc1, _, _, _, _, _, _, _, _ = lcrModel.main(
BASE_train + str(i) + '.txt', test + str(i) + '.txt',
accuracyOnt, test_size[i], remaining_size)
acc.append(acc1)
tf.reset_default_graph()
print('iteration: ' + str(i))
with open(
"cross_results_" + str(FLAGS.year) + "/LCRROT_" +
str(FLAGS.year) + '.txt', 'w') as result:
result.write(str(acc) + '\n')
result.write('Accuracy: {}, St Dev:{} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
if runLCRROTINVERSE == True:
acc = []
#k-fold cross validation
for i in range(split_size):
acc1, _, _, _, _, _ = lcrModelInverse.main(
BASE_train + str(i) + '.txt', test + str(i) + '.txt',
accuracyOnt, test_size[i], remaining_size)
acc.append(acc1)
tf.reset_default_graph()
print('iteration: ' + str(i))
with open(
"cross_results_" + str(FLAGS.year) + "/LCRROT_INVERSE_" +
str(FLAGS.year) + '.txt', 'w') as result:
result.write(str(acc))
result.write('Accuracy: {}, St Dev:{} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
if runLCRROTALT == True:
acc = []
#k-fold cross validation
for i in range(split_size):
acc1, _, _, _, _, _ = lcrModelAlt_hierarchical_v3.main(
BASE_train + str(i) + '.txt', test + str(i) + '.txt',
accuracyOnt, test_size[i], remaining_size)
acc.append(acc1)
tf.reset_default_graph()
print('iteration: ' + str(i))
with open(
FLAGS.hardcoded_path +
"/data/programGeneratedData/crossValidation" +
str(FLAGS.year) +
'/cross_results_"+str(FLAGS.year)+"/LCRROT_ALT_"+str(FLAGS.year)'
+ '.txt', 'w') as result:
result.write(str(acc))
result.write('Accuracy: {}, St Dev:{} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
if runCABASC == True:
acc = []
#k-fold cross validation
for i in range(split_size):
acc1, _, _ = cabascModel.main(BASE_train + str(i) + '.txt',
REMAIN_val + str(i) + '.txt',
accuracyOnt, test_size[i],
remaining_size)
acc.append(acc1)
tf.reset_default_graph()
print('iteration: ' + str(i))
with open(
"cross_results_" + str(FLAGS.year) + "/CABASC_" +
str(FLAGS.year) + '.txt', 'w') as result:
result.write(str(acc))
result.write('Accuracy: {}, St Dev:{} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
if runSVM == True:
acc = []
#k-fold cross validation
for i in range(split_size):
acc1 = svmModel.main(BASE_svm_train + str(i) + '.txt',
test + str(i) + '.txt', accuracyOnt,
test_size[i], remaining_size)
acc.append(acc1)
tf.reset_default_graph()
with open(
"cross_results_" + str(FLAGS.year) + "/SVM_" +
str(FLAGS.year) + '.txt', 'w') as result:
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {:.5f}, St Dev:{:.4f}'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
result.write(str(acc))
result.write('Accuracy: {:.5f}, St Dev:{:.4f} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print('Finished program succesfully')
if runLCRModelAlt_hierarchical_v4 == True:
print('Running CrossVal V4, year = ' + str(FLAGS.year))
acc = []
# k-fold cross validation
for i in range(split_size):
acc1 = lcrModelAlt_hierarchical_v4.main(
BASE_train + str(i) + '.txt', test + str(i) + '.txt',
accuracyOnt, test_size[i], remaining_size)
acc.append(acc1)
tf.reset_default_graph()
with open(
FLAGS.hardcoded_path +
"/data/programGeneratedData/crossValidation" +
str(FLAGS.year) + "cross_results_" + str(FLAGS.year) + "/v4_" +
str(FLAGS.year) + '.txt', 'w') as result:
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {:.5f}, St Dev:{:.4f}'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
result.write(str(acc))
result.write('Accuracy: {:.5f}, St Dev:{:.4f} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print('Finished program succesfully')
if runAdversarial == True:
print('Running CrossVal adversarial, year = ' + str(FLAGS.year))
acc = []
# k-fold cross validation
for i in range(split_size):
if FLAGS.year == 2015:
acc1, pred2, fw2, bw2, tl2, tr2 = adversarial.main(
BASE_train + str(i) + '.txt',
test + str(i) + '.txt',
accuracyOnt,
test_size[i],
remaining_size,
learning_rate_dis=0.02,
learning_rate_gen=0.002,
keep_prob=0.3,
momentum_dis=0.9,
momentum_gen=0.36,
l2=0.00001,
k=3,
WriteFile=False)
else:
acc1, pred2, fw2, bw2, tl2, tr2 = adversarial.main(
BASE_train + str(i) + '.txt',
test + str(i) + '.txt',
accuracyOnt,
test_size[i],
remaining_size,
learning_rate_dis=0.03,
learning_rate_gen=0.0045,
keep_prob=0.3,
momentum_dis=0.7,
momentum_gen=0.42,
l2=0.00001,
k=3,
WriteFile=False)
acc.append(acc1)
tf.reset_default_graph()
with open(
FLAGS.hardcoded_path +
"/data/programGeneratedData/crossValidation" +
str(FLAGS.year) + "cross_results_" + str(FLAGS.year) +
"/Adv_" + str(FLAGS.year) + '.txt', 'w') as result:
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {:.5f}, St Dev:{:.4f}'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
result.write(str(acc))
result.write('Accuracy: {:.5f}, St Dev:{:.4f} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print('Finished program succesfully')
def main(_):
loadData = True # only for non-contextualised word embeddings.
augment_data = False # Load data must be true to augment
useOntology = False
runLCRROTALT = True
#determine if backupmethod is used
if runLCRROTALT:
backup = True
else:
backup = False
BASE_train = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/cross_train_'
BASE_val = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/cross_val_'
BASE_svm_train = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/svm/cross_train_svm_'
BASE_svm_val = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/svm/cross_val_svm_'
REMAIN_val = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/cross_val_remainder_'
REMAIN_svm_val = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/svm/cross_val_remainder_'
# Number of k-fold cross validations
split_size = 10
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector = loadCrossValidation(
FLAGS, split_size, augment_data, loadData)
remaining_size = 248
accuracyOnt = 0.87
if useOntology == True:
print('Starting Ontology Reasoner')
acc = []
remaining_size_vec = []
#k-fold cross validation
for i in range(split_size):
Ontology = OntReasoner()
accuracyOnt, remaining_size = Ontology.run(
backup, BASE_val + str(i) + '.txt', runSVM, True, i)
acc.append(accuracyOnt)
remaining_size_vec.append(remaining_size)
with open(
"cross_results_" + str(FLAGS.year) + "/ONTOLOGY_" +
str(FLAGS.year) + '.txt', 'w') as result:
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
print(acc)
result.write('size:' + str(test_size))
result.write('accuracy: ' + str(acc) + '\n')
result.write('remaining size: ' + str(remaining_size_vec) + '\n')
result.write('Accuracy: {}, St Dev:{} \n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
test = REMAIN_val
else:
test = BASE_val
if runLCRROTALT == True:
acc = []
#k-fold cross validation
for i in range(split_size):
acc1, _, _, _, _, _ = lcrModelAlt.main(
BASE_train + str(i) + '.txt',
test + str(i) + '.txt',
accuracyOnt,
test_size[i],
remaining_size,
augment_data,
FLAGS.augmentation_file_path,
ct={})
acc.append(acc1)
tf.reset_default_graph()
print('iteration: ' + str(i))
with open(
"cross_results_" + str(FLAGS.year) + "/LCRROT_ALT_" +
str(FLAGS.year) + '.txt', 'w') as result:
result.write(str(acc))
result.write('Accuracy: {}, St Dev:{} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
print('Finished program succesfully')
def main(_):
loadData = False
# loadData = True
# useOntology = True
useOntology = False
runCABASC = False
runLCRROT = False
runLCRROTINVERSE = False
runLCRROTALT = True
runSVM = False
weightanalysis = False
#determine if backupmethod is used
if runCABASC or runLCRROT or runLCRROTALT or runLCRROTINVERSE or runSVM:
backup = True
else:
backup = False
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector = loadDataAndEmbeddings(
FLAGS, loadData)
# print(test_size)
remaining_size = 250
accuracyOnt = 0.87
if useOntology == True:
print('Starting Ontology Reasoner')
Ontology = OntReasoner()
#out of sample accuracy
accuracyOnt, remaining_size = Ontology.run(backup, FLAGS.test_path,
runSVM)
#in sample accuracy
Ontology = OntReasoner()
accuracyInSampleOnt, remaining_size = Ontology.run(
backup, FLAGS.train_path, runSVM)
if runSVM == True:
test = FLAGS.remaining_svm_test_path
else:
test = FLAGS.remaining_test_path
print('train acc = {:.4f}, test acc={:.4f}, remaining size={}'.format(
accuracyOnt, accuracyOnt, remaining_size))
else:
if runSVM == True:
test = FLAGS.test_svm_path
else:
test = FLAGS.test_path
# LCR-Rot model
if runLCRROT == True:
_, pred1, fw1, bw1, tl1, tr1, sent, target, true = lcrModel.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
# LCR-Rot-inv model
if runLCRROTINVERSE == True:
lcrModelInverse.main(FLAGS.train_path, test, accuracyOnt, test_size,
remaining_size)
tf.reset_default_graph()
# LCR-Rot-hop model
if runLCRROTALT == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
# CABASC model
if runCABASC == True:
_, pred3, weights = cabascModel.main(FLAGS.train_path, test,
accuracyOnt, test_size,
remaining_size)
if weightanalysis and runLCRROT and runLCRROTALT:
outF = open('sentence_analysis.txt', "w")
dif = np.subtract(pred3, pred1)
for i, value in enumerate(pred3):
if value == 1 and pred2[i] == 0:
sentleft, sentright = [], []
flag = True
for word in sent[i]:
if word == '$t$':
flag = False
continue
if flag:
sentleft.append(word)
else:
sentright.append(word)
print(i)
outF.write(str(i))
outF.write("\n")
outF.write(
'lcr pred: {}; CABASC pred: {}; lcralt pred: {}; true: {}'
.format(pred1[i], pred3[i], pred2[i], true[i]))
outF.write("\n")
outF.write(";".join(sentleft))
outF.write("\n")
outF.write(";".join(str(x) for x in fw1[i][0]))
outF.write("\n")
outF.write(";".join(sentright))
outF.write("\n")
outF.write(";".join(str(x) for x in bw1[i][0]))
outF.write("\n")
outF.write(";".join(target[i]))
outF.write("\n")
outF.write(";".join(str(x) for x in tl1[i][0]))
outF.write("\n")
outF.write(";".join(str(x) for x in tr1[i][0]))
outF.write("\n")
outF.write(";".join(sentleft))
outF.write("\n")
outF.write(";".join(str(x) for x in fw2[i][0]))
outF.write("\n")
outF.write(";".join(sentright))
outF.write("\n")
outF.write(";".join(str(x) for x in bw2[i][0]))
outF.write("\n")
outF.write(";".join(target[i]))
outF.write("\n")
outF.write(";".join(str(x) for x in tl2[i][0]))
outF.write("\n")
outF.write(";".join(str(x) for x in tr2[i][0]))
outF.write("\n")
outF.write(";".join(sent[i]))
outF.write("\n")
outF.write(";".join(str(x) for x in weights[i][0]))
outF.write("\n")
outF.close()
# BoW model
if runSVM == True:
svmModel.main(FLAGS.train_svm_path, test, accuracyOnt, test_size,
remaining_size)
print('Finished program succesfully')
def main(_):
loadData = False # only for non-contextualised word embeddings.
# Use prepareBERT for BERT (and BERT_Large) and prepareELMo for ELMo
useOntology = False # When run together with runLCRROTALT, the two-step method is used
runLCRROTALT = False
runSVM = False
runCABASC = False
runLCRROT = False
runLCRROTINVERSE = False
weightanalysis = False
runLCRROTALT_v1 = False
runLCRROTALT_v2 = False
runLCRROTALT_v3 = False
runLCRROTALT_v4 = True
#curriculum_learning = True
# if curriculum_learning = True, then choose either one_pass or baby_steps to be True as well!
runOne_Pass = False
runBaby_Steps = True
if runOne_Pass or runBaby_Steps: # if baby steps or one pass, then automatically curriculum learning True as well to get the sorted indices.
curriculum_learning = True
# determine if backupmethod is used
if runCABASC or runLCRROT or runLCRROTALT or runLCRROTINVERSE or runSVM or runLCRROTALT_v1 or runLCRROTALT_v2 or runLCRROTALT_v3 or runLCRROTALT_v4:
backup = True
else:
backup = False
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector = loadDataAndEmbeddings(FLAGS, loadData)
print(test_size)
remaining_size = 250
accuracyOnt = 0.87
tf.reset_default_graph()
if useOntology == True:
print('Starting Ontology Reasoner')
# in sample accuracy
Ontology = OntReasoner()
accuracyOnt, remaining_size = Ontology.run(backup, FLAGS.test_path_ont, runSVM)
# out of sample accuracy
# Ontology = OntReasoner()
# accuracyInSampleOnt, remainingInSample_size = Ontology.run(backup,FLAGS.train_path_ont, runSVM)
if runSVM == True:
test = FLAGS.remaining_svm_test_path
else:
test = FLAGS.remaining_test_path
print(test[0])
print('train acc = {:.4f}, test acc={:.4f}, remaining size={}'.format(accuracyOnt, accuracyOnt, remaining_size))
else:
if runSVM == True:
test = FLAGS.test_svm_path
else:
test = FLAGS.test_path
# Get curriculum learning scores, either the ones already saved, or new ones
# Make sure that the instances in FLAGS.train_path_ont and FLAGS.train_path (and the two test sets) have the same order of their instances!
if curriculum_learning == True:
try:
sort_ind = pickle.load(open(FLAGS.sorted_indices, "rb"))
except:
tr_features, tr_sent = sentiWordNet.main(FLAGS.train_path_ont, FLAGS.train_aspect_categories)
te_features, te_sent = sentiWordNet.main(FLAGS.test_path_ont, FLAGS.test_aspect_categories)
tr_sent = np.asarray(utils.change_y_to_onehot(tr_sent))
te_sent = np.asarray(utils.change_y_to_onehot(te_sent))
print(tr_features.shape)
print(tr_sent.shape)
print(te_features.shape)
print(te_sent.shape)
curr_scores = auxModel.main(tr_features, te_features, tr_sent, te_sent)
tf.reset_default_graph()
inds1 = np.arange(0, len(curr_scores))
sort_ind = [x for _, x in sorted(zip(curr_scores, inds1))]
pickle.dump(sort_ind, open(FLAGS.sorted_indices, "wb"))
# LCR-Rot-hop model
if runLCRROTALT == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt.main(FLAGS.train_path, test, accuracyOnt, test_size,
remaining_size)
tf.reset_default_graph()
if runLCRROTALT_v1 == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt_hierarchical_v1.main(FLAGS.train_path, test, accuracyOnt, test_size,
remaining_size)
tf.reset_default_graph()
if runLCRROTALT_v2 == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt_hierarchical_v2.main(FLAGS.train_path, test, accuracyOnt, test_size,
remaining_size)
tf.reset_default_graph()
if runLCRROTALT_v3 == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt_hierarchical_v3.main(FLAGS.train_path, test, accuracyOnt, test_size,
remaining_size)
tf.reset_default_graph()
if runLCRROTALT_v4 == True:
if runOne_Pass:
acc = lcrModelAlt_hierarchical_v4_one_pass.main(FLAGS.train_path, test,
accuracyOnt,
test_size,
remaining_size,
sort_ind, FLAGS.num_buckets)
tf.reset_default_graph()
elif runBaby_Steps == True:
tf.reset_default_graph()
acc = lcrModelAlt_hierarchical_v4_baby_steps.main(FLAGS.train_path, test, accuracyOnt,
test_size,
remaining_size,
sort_ind, FLAGS.num_buckets)
tf.reset_default_graph()
else:
acc = lcrModelAlt_hierarchical_v4_trainevaltest.main(FLAGS.hyper_train_path, FLAGS.hyper_eval_path, test, FLAGS.train_path,
accuracyOnt,
test_size,
remaining_size)
tf.reset_default_graph()
def main(_):
loadData = False
useOntology = False
runCABASC = False
runLCRROT = False
runLCRROTINVERSE = False
runLCRROTALT = False
runSVM = False
runlcrDoubleRAA = True
runINVMULTIHOP1 = False
runlcrDoubleRAAtype2 = False
weightanalysis = False
#determine if backupmethod is used
if runCABASC or runLCRROT or runLCRROTALT or runLCRROTINVERSE or runSVM or runlcrDoubleRAA or runINVMULTIHOP1:
backup = True
else:
backup = False
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector = loadDataAndEmbeddings(
FLAGS, loadData)
print(test_size)
remaining_size = 250
accuracyOnt = 0.87
if useOntology == True:
print('Starting Ontology Reasoner')
Ontology = OntReasoner()
#out of sample accuracy
accuracyOnt, remaining_size = Ontology.run(backup, FLAGS.test_path,
runSVM)
#in sample accuracy
Ontology = OntReasoner()
accuracyInSampleOnt, remaining_size = Ontology.run(
backup, FLAGS.train_path, runSVM)
if runSVM == True:
test = FLAGS.remaining_svm_test_path
else:
test = FLAGS.remaining_test_path
print('train acc = {:.4f}, test acc={:.4f}, remaining size={}'.format(
accuracyOnt, accuracyOnt, remaining_size))
else:
if runSVM == True:
test = FLAGS.test_svm_path
else:
test = FLAGS.test_path
# LCR-Rot model
if runLCRROT == True:
_, pred1, fw1, bw1, tl1, tr1, sent, target, true = lcrModel.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
# LCR-Rot-inv model
if runLCRROTINVERSE == True:
lcrModelInverse.main(FLAGS.train_path, test, accuracyOnt, test_size,
remaining_size)
tf.reset_default_graph()
# LCR-Rot-hop model
if runLCRROTALT == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
if runlcrDoubleRAA == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrDoubleRAA.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
if runINVMULTIHOP1 == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrinvmodel2.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
if runlcrDoubleRAAtype2 == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrDoubleRAAtype2.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
# BoW model
if runSVM == True:
svmModel.main(FLAGS.train_svm_path, test, accuracyOnt, test_size,
remaining_size)
print('Finished program succesfully')
def main(_):
loadData = False
useOntology = False
runCABASC = False
runLCRROT = False
runLCRROTINVERSE = False
runLCRROTALT = True
runSVM = False
#determine if backupmethod is used
if runCABASC or runLCRROT or runLCRROTALT or runLCRROTINVERSE or runSVM:
backup = True
else:
backup = False
BASE_train = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/cross_train_'
BASE_val = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/cross_val_'
BASE_svm_train = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/svm/cross_train_svm_'
BASE_svm_val = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/svm/cross_val_svm_'
REMAIN_val = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/cross_val_remainder_'
REMAIN_svm_val = "data/programGeneratedData/crossValidation" + str(
FLAGS.year) + '/svm/cross_val_remainder_'
# Number of k-fold cross validations
split_size = 10
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector = loadCrossValidation(
FLAGS, split_size, loadData)
remaining_size = 248
accuracyOnt = 0.87
if useOntology == True:
print('Starting Ontology Reasoner')
acc = []
remaining_size_vec = []
#k-fold cross validation
for i in range(split_size):
Ontology = OntReasoner()
accuracyOnt, remaining_size = Ontology.run(
backup, BASE_val + str(i) + '.txt', runSVM, True, i)
acc.append(accuracyOnt)
remaining_size_vec.append(remaining_size)
with open(
"cross_results_" + str(FLAGS.year) + "/ONTOLOGY_" +
str(FLAGS.year) + '.txt', 'w') as result:
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
print(acc)
result.write('size:' + str(test_size))
result.write('accuracy: ' + str(acc) + '\n')
result.write('remaining size: ' + str(remaining_size_vec) + '\n')
result.write('Accuracy: {}, St Dev:{} \n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
if runSVM == True:
test = REMAIN_svm_val
else:
test = REMAIN_val
else:
if runSVM == True:
test = BASE_svm_val
else:
test = BASE_val
if runLCRROT == True:
acc = []
#k-fold cross validation
for i in [8]:
acc1, _, _, _, _, _, _, _, _ = lcrModel.main(
BASE_train + str(i) + '.txt', test + str(i) + '.txt',
accuracyOnt, test_size[i], remaining_size)
acc.append(acc1)
tf.reset_default_graph()
print('iteration: ' + str(i))
with open(
"cross_results_" + str(FLAGS.year) + "/LCRROT_" +
str(FLAGS.year) + '.txt', 'w') as result:
result.write(str(acc) + '\n')
result.write('Accuracy: {}, St Dev:{} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
if runLCRROTINVERSE == True:
acc = []
#k-fold cross validation
for i in range(split_size):
acc1, _, _, _, _, _ = lcrModelInverse.main(
BASE_train + str(i) + '.txt', test + str(i) + '.txt',
accuracyOnt, test_size[i], remaining_size)
acc.append(acc1)
tf.reset_default_graph()
print('iteration: ' + str(i))
with open(
"cross_results_" + str(FLAGS.year) + "/LCRROT_INVERSE_" +
str(FLAGS.year) + '.txt', 'w') as result:
result.write(str(acc))
result.write('Accuracy: {}, St Dev:{} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
if runLCRROTALT == True:
acc = []
#k-fold cross validation
for i in range(split_size):
acc1, _, _, _, _, _ = lcrModelAlt.main(
BASE_train + str(i) + '.txt', test + str(i) + '.txt',
accuracyOnt, test_size[i], remaining_size)
acc.append(acc1)
tf.reset_default_graph()
print('iteration: ' + str(i))
with open(
"cross_results_" + str(FLAGS.year) + "/LCRROT_ALT_" +
str(FLAGS.year) + '.txt', 'w') as result:
result.write(str(acc))
result.write('Accuracy: {}, St Dev:{} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
if runCABASC == True:
acc = []
#k-fold cross validation
for i in range(split_size):
acc1, _, _ = cabascModel.main(BASE_train + str(i) + '.txt',
REMAIN_val + str(i) + '.txt',
accuracyOnt, test_size[i],
remaining_size)
acc.append(acc1)
tf.reset_default_graph()
print('iteration: ' + str(i))
with open(
"cross_results_" + str(FLAGS.year) + "/CABASC_" +
str(FLAGS.year) + '.txt', 'w') as result:
result.write(str(acc))
result.write('Accuracy: {}, St Dev:{} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {}, St Dev:{}'.format(np.mean(np.asarray(acc)),
np.std(np.asarray(acc))))
if runSVM == True:
acc = []
#k-fold cross validation
for i in range(split_size):
acc1 = svmModel.main(BASE_svm_train + str(i) + '.txt',
test + str(i) + '.txt', accuracyOnt,
test_size[i], remaining_size)
acc.append(acc1)
tf.reset_default_graph()
with open(
"cross_results_" + str(FLAGS.year) + "/SVM_" +
str(FLAGS.year) + '.txt', 'w') as result:
print(str(split_size) + '-fold cross validation results')
print('Accuracy: {:.5f}, St Dev:{:.4f}'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
result.write(str(acc))
result.write('Accuracy: {:.5f}, St Dev:{:.4f} /n'.format(
np.mean(np.asarray(acc)), np.std(np.asarray(acc))))
print('Finished program succesfully')
def main(_):
loadData = False # only for non-contextualised word embeddings.
# Use prepareBERT for BERT (and BERT_Large) and prepareELMo for ELMo
useOntology = False # When run together with runLCRROTALT, the two-step method is used
runLCRROTALT = False
runSVM = False
runCABASC = False
runLCRROT = False
runLCRROTINVERSE = False
weightanalysis = False
runLCRROTALT_v1 = False
runLCRROTALT_v2 = False
runLCRROTALT_v3 = False
runLCRROTALT_v4 = True
runAdversarial = False
#Save and Restore if desired
Save = False
RestoreSave = False
restore_path = '/Users/ronhochstenbach/Desktop/Ectrie Thesis/Venv_Thesis/Saved_Models/2020-05-29 23:08:39.394204_BERT_2016/Iter_0-470' #do not add .meta!
# determine if backupmethod is used
if runCABASC or runLCRROT or runLCRROTALT or runLCRROTINVERSE or runSVM or runLCRROTALT_v1 or runLCRROTALT_v2 or runLCRROTALT_v3 or runLCRROTALT_v4 or runAdversarial:
backup = True
else:
backup = False
# retrieve data and wordembeddings
train_size, test_size, train_polarity_vector, test_polarity_vector = loadDataAndEmbeddings(
FLAGS, loadData)
print(train_size)
print(test_size)
remaining_size = 250
accuracyOnt = 0.87
if useOntology == True:
print('Starting Ontology Reasoner')
# in sample accuracy
Ontology = OntReasoner()
accuracyOnt, remaining_size = Ontology.run(backup, FLAGS.test_path_ont,
runSVM)
# out of sample accuracy
# Ontology = OntReasoner()
# accuracyInSampleOnt, remainingInSample_size = Ontology.run(backup,FLAGS.train_path_ont, runSVM)
if runSVM == True:
test = FLAGS.remaining_svm_test_path
else:
test = FLAGS.remaining_test_path
print(test[0])
print('train acc = {:.4f}, test acc={:.4f}, remaining size={}'.format(
accuracyOnt, accuracyOnt, remaining_size))
else:
if runSVM == True:
test = FLAGS.test_svm_path
else:
test = FLAGS.test_path
# LCR-Rot-hop model
if runLCRROTALT == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
if runLCRROTALT_v1 == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt_hierarchical_v1.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
if runLCRROTALT_v2 == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt_hierarchical_v2.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
if runLCRROTALT_v3 == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt_hierarchical_v3.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
if runLCRROTALT_v4 == True:
_, pred2, fw2, bw2, tl2, tr2 = lcrModelAlt_hierarchical_v4.main(
FLAGS.train_path, test, accuracyOnt, test_size, remaining_size)
tf.reset_default_graph()
if runAdversarial == True:
print('Running Adversarial')
_, pred2, fw2, bw2, tl2, tr2 = adversarial.main(
FLAGS.train_path,
test,
accuracyOnt,
test_size,
remaining_size,
learning_rate_dis=0.02,
learning_rate_gen=0.002,
keep_prob=0.3,
momentum_dis=0.9,
momentum_gen=0.36,
l2=0.00001,
k=3,
WriteFile=True)
tf.reset_default_graph()