def evaluation(self):
if self.hyper.is_bert == "nyt_bert_tokenizer" or \
self.hyper.is_bert == "nyt11_bert_tokenizer" or \
self.hyper.is_bert == "nyt10_bert_tokenizer":
dev_set = Selection_bert_Nyt_Dataset(self.hyper, self.hyper.dev)
loader = Selection_bert_loader(dev_set,
batch_size=100,
pin_memory=True)
elif self.hyper.is_bert == "bert_bilstem_crf":
dev_set = Selection_Nyt_Dataset(self.hyper, self.hyper.dev)
loader = Selection_loader(dev_set, batch_size=100, pin_memory=True)
else:
dev_set = Selection_Dataset(self.hyper, self.hyper.dev)
loader = Selection_loader(dev_set, batch_size=100, pin_memory=True)
self.metrics.reset()
self.model.eval()
pbar = tqdm(enumerate(BackgroundGenerator(loader)), total=len(loader))
with torch.no_grad():
for batch_ndx, sample in pbar:
output = self.model(sample, is_train=False)
#self.metrics(output['selection_triplets'], output['spo_gold'])
self.metrics(output, output)
result = self.metrics.get_metric()
print(', '.join([
"%s: %.4f" % (name, value)
for name, value in result.items() if not name.startswith("_")
]) + " ||")
def train(self):
train_set = Selection_Dataset(self.hyper, self.hyper.train)
loader = Selection_loader(train_set,
batch_size=self.hyper.train_batch,
pin_memory=True)
for epoch in range(self.hyper.epoch_num):
self.model.train()
pbar = tqdm(enumerate(BackgroundGenerator(loader)),
total=len(loader))
for batch_idx, sample in pbar:
self.optimizer.zero_grad()
output = self.model(sample, is_train=True)
loss = output['loss']
loss.backward()
self.optimizer.step()
pbar.set_description(output['description'](
epoch, self.hyper.epoch_num))
# break
self.save_model(epoch)
if epoch % self.hyper.print_epoch == 0 and epoch > 3:
self.evaluation()
def evaluation(self):
dev_set = Selection_Dataset(self.hyper, self.hyper.dev)
loader = Selection_loader(dev_set,
batch_size=self.hyper.eval_batch,
pin_memory=True)
self.triplet_metrics.reset()
self.model.eval()
pbar = tqdm(enumerate(BackgroundGenerator(loader)), total=len(loader))
with torch.no_grad():
for batch_ndx, sample in pbar:
output = self.model(sample, is_train=False)
self.triplet_metrics(output['selection_triplets'],
output['spo_gold'])
self.ner_metrics(output['gold_tags'], output['decoded_tag'])
triplet_result = self.triplet_metrics.get_metric()
ner_result = self.ner_metrics.get_metric()
print('Triplets-> ' + ', '.join([
"%s: %.4f" % (name[0], value)
for name, value in triplet_result.items()
if not name.startswith("_")
]) + ' ||' + 'NER->' + ', '.join([
"%s: %.4f" % (name[0], value)
for name, value in ner_result.items()
if not name.startswith("_")
]))
def postcheck(self):
dev_set = Selection_Dataset(self.hyper,
self.hyper.xgb_train_root,
is_xgb=True)
loader = Selection_loader(dev_set,
batch_size=self.hyper.eval_batch,
pin_memory=True)
self.model.eval()
pbar = tqdm(enumerate(BackgroundGenerator(loader)), total=len(loader))
self.save_rc.reset('pos_neg_score.json')
flag = False
with torch.no_grad():
for batch_ndx, sample in pbar:
output = self.model(sample, is_train=False)
if output.get('pos_list'):
pos_list = output['pos_list']
flag = True
neg_list = output['neg_list']
if flag:
self.save_rc.save(neg_list, pos_list)
else:
self.save_rc.save(neg_list)
def evaluation(self):
dev_set = Selection_Dataset(self.hyper, self.hyper.dev)
loader = Selection_loader(dev_set,
batch_size=self.hyper.eval_batch,
pin_memory=True)
self.selection_metrics.reset()
self.model.eval()
pbar = tqdm(enumerate(BackgroundGenerator(loader)), total=len(loader))
with torch.no_grad():
for batch_ndx, sample in pbar:
pred = self.model(sample, is_train=False)
pred = torch.sigmoid(pred) > 0.5
labels = sample.selection_id
self.selection_metrics(
np.array(pred.cpu().numpy(), dtype=int).tolist(),
np.array(labels.cpu().numpy(), dtype=int).tolist())
triplet_result = self.selection_metrics.get_metric()
print('Triplets-> ' + ', '.join([
"%s: %.4f" % (name[0], value)
for name, value in triplet_result.items()
if not name.startswith("_")
]))
def train(self):
if self.hyper.is_bert == "bert_bilstem_crf":
train_set = Selection_Nyt_Dataset(self.hyper, self.hyper.train)
loader = Selection_loader(train_set,
batch_size=100,
pin_memory=True)
elif self.hyper.is_bert == "nyt_bert_tokenizer" or \
self.hyper.is_bert == "nyt11_bert_tokenizer" or \
self.hyper.is_bert == "nyt10_bert_tokenizer":
train_set = Selection_bert_Nyt_Dataset(self.hyper,
self.hyper.train)
loader = Selection_bert_loader(train_set,
batch_size=100,
pin_memory=True)
else:
train_set = Selection_Dataset(self.hyper, self.hyper.train)
loader = Selection_loader(train_set,
batch_size=100,
pin_memory=True)
for epoch in range(self.hyper.epoch_num):
self.model.train()
pbar = tqdm(enumerate(BackgroundGenerator(loader)),
total=len(loader))
for batch_idx, sample in pbar:
self.optimizer.zero_grad()
output = self.model(sample, is_train=True)
loss = output['loss']
loss.backward()
self.optimizer.step()
pbar.set_description(output['description'](
epoch, self.hyper.epoch_num))
self.save_model(epoch)
if epoch > 3:
self.evaluation()
#if epoch >= 6:
# self.evaluation()
'''
def evaluation(self):
dev_set = Selection_Dataset(self.hyper, self.hyper.dev)
loader = Selection_loader(dev_set,
batch_size=self.hyper.eval_batch,
pin_memory=True)
self.triplet_metrics.reset()
self.model.eval()
self.model._init_spo_search(SPO_searcher(self.hyper))
pbar = tqdm(enumerate(BackgroundGenerator(loader)), total=len(loader))
self.save_rc.reset('dev.json')
with torch.no_grad():
for batch_ndx, sample in pbar:
output = self.model(sample, is_train=False)
"""
attn = self.model.attn.cpu()
if batch_ndx == 41:
print(sample.text[0])
attn_multi_plot(attn,sample.text[0])
"""
# distant search
self.triplet_metrics(output['selection_triplets'],
output['spo_gold'])
self.ner_metrics(output['gold_tags'], output['decoded_tag'])
#self.save_err.save(batch_ndx,output['selection_triplets'], output['spo_gold'])
self.save_rc.save(output['selection_triplets'])
triplet_result = self.triplet_metrics.get_metric()
ner_result = self.ner_metrics.get_metric()
print('Triplets-> ' + ', '.join([
"%s: %.4f" % (name[0], value)
for name, value in triplet_result.items()
if not name.startswith("_")
]) + ' ||' + 'NER->' + ', '.join([
"%s: %.4f" % (name[0], value)
for name, value in ner_result.items()
if not name.startswith("_")
]))
def train_pso(self, epoch=None):
train_set = Selection_Dataset(self.hyper, self.hyper.train)
loader = Selection_loader(train_set,
batch_size=self.hyper.train_batch,
pin_memory=True)
if not epoch:
epoch = 0
num_train_steps = int(
len(loader) / self.hyper.train_batch *
(self.hyper.epoch_num - epoch + 1))
num_warmup_steps = int(num_train_steps * self.hyper.warmup_prop)
self.scheduler = self._scheduler(self.optimizer, num_warmup_steps,
num_train_steps)
while epoch <= self.hyper.epoch_num:
pbar = tqdm(enumerate(BackgroundGenerator(loader)),
total=len(loader))
for batch_idx, sample in pbar:
self.model.train()
output = self.model(sample, is_train=True)
loss = output['loss']
loss.backward()
torch.nn.utils.clip_grad_norm_(self.model.parameters(), 1.0)
self.optimizer.step()
self.scheduler.step()
self.model.zero_grad()
pbar.set_description(output['description'](
epoch, self.hyper.epoch_num))
self.save_model(epoch)
epoch += 1
def train(self, epoch=None):
train_set = Selection_Dataset(self.hyper, self.hyper.train)
#sampler = torch.utils.data.sampler.WeightedRandomSampler(train_set.weight,len(train_set.weight))
loader = Selection_loader(train_set,
batch_size=self.hyper.train_batch,
pin_memory=True)
#loader = Selection_loader(train_set, batch_size=self.hyper.train_batch,sampler = sampler, pin_memory=True)
if not epoch:
epoch = 0
while epoch <= self.hyper.epoch_num:
#for epoch in range(self.hyper.epoch_num):
self.model.train()
pbar = tqdm(enumerate(BackgroundGenerator(loader)),
total=len(loader))
for batch_idx, sample in pbar:
self.optimizer.zero_grad()
output = self.model(sample, is_train=True)
loss = output['loss']
loss.backward()
self.optimizer.step()
pbar.set_description(output['description'](
epoch, self.hyper.epoch_num))
self.save_model(epoch)
"""
if epoch % self.hyper.print_epoch == 0 and epoch > 3:
if self.model_name == 'selection':
self.evaluation()
elif self.model_name == 'pso_1':
self.evaluation_pso()
"""
epoch += 1
def evaluation(self):
dev_set = Selection_Dataset(self.hyper, self.hyper.dev)
loader = Selection_loader(dev_set,
batch_size=self.hyper.eval_batch,
pin_memory=True)
self.triplet_metrics.reset()
self.model.eval()
pbar = tqdm(enumerate(BackgroundGenerator(loader)), total=len(loader))
with torch.no_grad():
with open('./output/sub00.csv', 'w') as file:
id = 0
for batch_ndx, sample in pbar:
tokens = sample.tokens_id.to(self.device)
selection_gold = sample.selection_id.to(self.device)
bio_gold = sample.bio_id.to(self.device)
text_list = sample.text
spo_gold = sample.spo_gold
bio_text = sample.bio
output = self.model(sample, is_train=False)
self.triplet_metrics(output['selection_triplets'],
output['spo_gold'])
self.ner_metrics(output['gold_tags'],
output['decoded_tag'])
for i in range(len(output['decoded_tag'])):
file.write(str(8001 + id) + ',')
if len(output['selection_triplets'][i]) != 0:
file.write(output['selection_triplets'][i][0]
['predicate'] + ',')
file.write(
output['selection_triplets'][i][0]['subject'] +
',')
file.write(
output['selection_triplets'][i][0]['object'] +
'\n')
else:
if output['decoded_tag'][i].count('B') < 2:
file.write('Other' + ',' + 'Other' + ',' +
'Other')
else:
BIO = output['decoded_tag'][i]
tt = ''.join(reversed(BIO))
index1 = BIO.index('B')
index2 = len(tt) - tt.index('B') - 1
file.write('Other' + ',' +
text_list[i][index2] + ',' +
text_list[i][index1])
file.write('\n')
id += 1
# file.write('sentence {} BIO:\n'.format(i))
# for j in range(len(text_list[i])):
# file.write(text_list[i][j]+' ')
# file.write('\n')
# file.writelines(bio_text[i])
# file.write('\n')
#
# file.writelines(output['decoded_tag'][i])
# file.write('\n')
# file.writelines(output['gold_tags'][i])
# file.write('\n')
# file.write('sentence {} relation:\n'.format(i))
# file.write('\n')
# if len(output['selection_triplets']) == 0:
# file.write('empty')
# else:
# file.writelines(str(output['selection_triplets'][i]))
# file.write('\n')
# file.writelines(str(output['spo_gold'][i]))
# file.write('\n')
triplet_result = self.triplet_metrics.get_metric()
ner_result = self.ner_metrics.get_metric()
# print('triplet_result=', triplet_result)
# print('ner_result=', ner_result)
print('Triplets-> ' + ', '.join([
"%s: %.4f" % (name[0], value)
for name, value in triplet_result.items()
if not name.startswith("_")
]) + ' ||' + 'NER->' + ', '.join([
"%s: %.4f" % (name[0], value)
for name, value in ner_result.items()
if not name.startswith("_")
]))
def evaluation_pso(self):
dev_set = Selection_Dataset(self.hyper, self.hyper.dev)
loader = Selection_loader(dev_set,
batch_size=self.hyper.eval_batch,
pin_memory=True)
self.p_metrics.reset()
self.triplet_metrics.reset()
self.model.eval()
if self.model_name == 'pso_2':
self.model_p.eval()
all_labels, all_logits = [], []
self.model._init_spo_search(SPO_searcher(self.hyper))
pbar = tqdm(enumerate(BackgroundGenerator(loader)), total=len(loader))
if self.model_name == 'pso_1':
with torch.no_grad():
for batch_ndx, sample in pbar:
output = self.model(sample, is_train=False)
"""
if batch_ndx == 44:
attn = self.model.attn.cpu()
print(attn)
attn_plot(attn,sample.text[0])
"""
self.p_metrics(output['p_decode'], output['p_golden'])
all_labels.extend(output['p_golden'])
all_logits.extend(output['p_logits'])
"""
with open('labels.txt','w',encoding = 'utf-8') as f:
for l in all_labels:
f.write(' '.join(l))
f.write('\n')
with open('scores.txt','w',encoding ='utf-8') as t:
for l in all_logits:
t.write(' '.join([str(i) for i in l]))
t.write('\n')
"""
p_result = self.p_metrics.get_metric()
print('P-> ' + ', '.join([
"%s: %.4f" % (name[0], value)
for name, value in p_result.items()
if not name.startswith("_")
]))
#roc_auc_class(all_labels,all_logits)
elif self.model_name == 'pso_2':
self.save_rc.reset('dev.json')
with torch.no_grad():
for batch_ndx, sample in pbar:
output_p = self.model_p(sample, is_train=False)
self.p_metrics(output_p['p_decode'], output_p['p_golden'])
all_labels.extend(output_p['p_decode'])
all_logits.extend(output_p['p_golden'])
dev_set.schema_transformer(output_p, sample)
output = self.model(sample, is_train=False)
"""
if batch_ndx == 63:
print(sample.text[0])
attn = self.model.attn[2].cpu()
attn_pso_plot_sub(attn,sample.text[0])
for id in range(9,12):
attn = self.model.attn[id-9].cpu()
attn_pso_plot(attn,sample.text[0],id)
#attn_pso_plot_stack(attn,sample.text[0],id)
attn = self.model.attn[-1].cpu()
attn_pso_plot_sub(attn,sample.text[0])
"""
self.triplet_metrics(output['selection_triplets'],
output['spo_gold'])
self.save_err.save(batch_ndx, output['selection_triplets'],
output['spo_gold'])
self.save_rc.save(output['selection_triplets'])
p_result = self.p_metrics.get_metric()
triplet_result = self.triplet_metrics.get_metric()
print('Triplets-> ' + ', '.join([
"%s: %.4f" % (name[0], value)
for name, value in triplet_result.items()
if not name.startswith("_")
]))
print('P-> ' + ', '.join([
"%s: %.4f" % (name[0], value)
for name, value in p_result.items()
if not name.startswith("_")
]))