except:
step = 0
test_abs(args, device_id, cp, step)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step)
elif (args.task == 'ext'):
if (args.mode == 'train'):
train_ext(args, device_id)
elif (args.mode == 'validate'):
validate_ext(args, device_id)
if (args.mode == 'test'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_ext(args, device_id, cp, step)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step)
elif args.mode == 'test':
checkpoint = args.load_model
# Print or save summaries and probas for test mode
logging.info("Processing files...")
with open('../results/patents_analysis.csv', mode='w') as file:
writer = csv.writer(file,
delimiter='?',
quotechar='"',
quoting=csv.QUOTE_MINIMAL)
writer.writerow([
'Ref patent', 'Domain', 'Probability of contradiction',
'First part Contradiction', 'Second part Contradiction',
'Context'
])
for ref_patents, summaries, output_probas, prediction_contradiction, str_context in test_ext(
args, device_id, checkpoint, 0):
for ref_patent, summary, output_proba, prediction_c, context in zip(
ref_patents, summaries, output_probas,
prediction_contradiction, str_context):
# print(summary)
# Get patents domain if possible
try:
with open(
'../data_patents/input_data/test_directory/' +
ref_patent + '/' + ref_patent + '.DOMAIN',
mode='r') as domain_f:
name_domain = domain_f.read()
except:
name_domain = 'unknown'
step = 0
test_abs(args, device_id, cp, step)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step)
elif (args.task == 'ext'):
if (args.mode == 'train'):
# print('joint: ' + str(args.section_prediction))
train_ext(args, device_id, args.section_prediction)
elif (args.mode == 'validate'):
validate_ext(args, device_id)
if (args.mode == 'test'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_ext(args, device_id, cp, step, args.section_prediction)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step)
def run(args, hpspace):
if (args.task == 'abs'):
if (args.mode == 'train'):
if (args.hyperopt):
t = time.time()
newT = str(t).split(".")
args.model_path = "../models/" + newT[0]
args.log_file = "../logs/abs_bert_abs_" + newT[0]
args.lr_bert = hpspace['lr_bert']
args.lr_dec = hpspace['lr_dec']
args.accum_count = int(hpspace['accum_count'])
args.beta1 = hpspace['beta1']
args.beta2 = hpspace['beta2']
args.visible_gpus = '0'
args.bert_model = '..temp/bert-base-danish-uncased-v2'
args.vocab = '..temp/bert-base-danish-uncased-v2'
train_stats = train_abs(args, device_id)
x = train_stats.x
ppl = train_stats.perplexity
acc = train_stats.acc
print(x)
return {
'loss': x,
'eval_time': time.time(),
'status': STATUS_OK,
'other_stuff': {
'ppl': ppl,
'acc': acc
}
}
elif (args.mode == 'validate'):
validate_abs(args, device_id)
elif (args.mode == 'lead'):
baseline(args, cal_lead=True)
elif (args.mode == 'oracle'):
baseline(args, cal_oracle=True)
if (args.mode == 'test'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_abs(args, device_id, cp, step)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step)
elif (args.task == 'ext'):
if (args.mode == 'train'):
train_ext(args, device_id)
elif (args.mode == 'validate'):
validate_ext(args, device_id)
if (args.mode == 'test'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_ext(args, device_id, cp, step)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step)
if (args.mode == "sent_label"):
step = 0
sent_label_ext(args, device_id)