dev_datasets = [('aida-A', conll.testA), ('aida-B', conll.testB),
('msnbc', conll.msnbc), ('aquaint', conll.aquaint),
('ace2004', conll.ace2004), ('clueweb', conll.clueweb),
('wikipedia', conll.wikipedia)]
if args.mode == 'train':
print('training...')
config = {'lr': args.learning_rate, 'n_epochs': args.n_epochs}
pprint(config)
ranker.train(conll.train, dev_datasets, config)
elif args.mode == 'eval':
org_dev_datasets = dev_datasets # + [('aida-train', conll.train)]
dev_datasets = []
for dname, data in org_dev_datasets:
dev_datasets.append(
(dname, ranker.get_data_items(data, predict=True)))
print(dname, '#dev docs', len(dev_datasets[-1][1]))
vecs = ranker.model.rel_embs.cpu().data.numpy()
for di, (dname, data) in enumerate(dev_datasets):
ranker.model._coh_ctx_vecs = []
# predict each dataset one by one
predictions = ranker.predict(data)
print(
dname,
utils.tokgreen(
'micro F1: ' +
str(D.eval(org_dev_datasets[di][1], predictions))))
def train(self,
org_train_dataset,
org_dev_datasets,
config,
preranked_train=None,
preranked_dev=None):
print('extracting training data')
if preranked_train is None:
train_dataset = self.get_data_items(org_train_dataset,
predict=False)
else:
train_dataset = preranked_train
print('#train docs', len(train_dataset))
if preranked_dev is None:
dev_datasets = []
for dname, data in org_dev_datasets:
dev_datasets.append(
(dname, self.get_data_items(data, predict=True)))
print(dname, '#dev docs', len(dev_datasets[-1][1]))
else:
dev_datasets = preranked_dev
print('creating optimizer')
optimizer = optim.Adam(
[p for p in self.model.parameters() if p.requires_grad],
lr=config['lr'])
best_f1 = -1
not_better_count = 0
is_counting = False
stop = False
eval_after_n_epochs = self.args.eval_after_n_epochs
final_result_str = ''
print('total training items', len(train_dataset))
n_updates = 0
if config['multi_instance']:
n_updates_to_eval = 1000
n_updates_to_stop = 60000
f1_threshold = 0.875
f1_start_couting = 0.87
elif config['semisup']:
n_updates_to_eval = 5000
n_update_to_stop = 1e10
f1_threshold = 0.86
f1_start_couting = 0.86
else: # for supervised learning
n_updates_to_eval = 1000
n_updates_to_stop = 1000 * self.args.n_epochs
f1_threshold = 0.95
f1_start_couting = 0.95
for e in range(config['n_epochs']):
shuffle(train_dataset)
total_loss = 0
total = 0
for dc, batch in enumerate(
train_dataset): # each document is a minibatch
self.model.train()
optimizer.zero_grad()
tps = [m['selected_cands']['true_pos'] >= 0 for m in batch]
any_true = np.any(tps)
if any_true:
inputs = self.minibatch2input(batch)
else:
inputs = self.minibatch2input(batch, topk=2)
if config['semisup']:
if any_true: # from supervision (i.e. CoNLL)
scores = self.model.forward(
inputs,
gold=inputs['true_pos'].view(-1, 1),
inference='LBP')
else:
scores = self.model.forward(
inputs,
gold=inputs['true_pos'].view(-1, 1),
inference='star')
else:
scores = self.model.forward(inputs,
gold=inputs['true_pos'].view(
-1, 1))
if any_true:
loss = self.model.loss(scores, inputs['true_pos'])
else:
loss = self.model.multi_instance_loss(scores, inputs)
loss.backward()
optimizer.step()
loss = loss.cpu().data.item()
total_loss += loss
if dc % 100 == 0:
print('epoch',
e,
"%0.2f%%" % (dc / len(train_dataset) * 100),
loss,
end='\r')
n_updates += 1
if n_updates % n_updates_to_eval == 0:
# only continue if the best f1 is larger than
if n_updates >= n_updates_to_stop and best_f1 < f1_threshold:
stop = True
print(
'this initialization is not good. Run another one... STOP'
)
break
print('\n--------------------')
dev_f1 = 0
results = ''
for di, (dname, data) in enumerate(dev_datasets):
#if dname == 'aida-A': #dname != '':
if dname != '':
#a = 0.1
#b = 1.
#c = 0.95
#global alpha
#global beta
#global gamma
#alpha = a
#beta = b
#gamma = c
predictions = self.predict(data, n_best=n_best)
cats = None
# **YD** only ignore .conll for reddit data
# if'cat' in data[0][0]['raw']['conll_m']:
if 'reddit' not in dname and 'cat' in data[0][0][
'raw']['conll_m']:
cats = []
for doc in data:
cats += [
m['raw']['conll_m']['cat'] for m in doc
]
# **YD** change output of D.eval to include prec, rec and f1
"""
if cats is None:
f1 = D.eval(org_dev_datasets[di][1], predictions)
else:
f1 = D.eval(org_dev_datasets[di][1], predictions, cats)
#print(alpha, beta, gamma, dname, utils.tokgreen('micro F1: ' + str(f1)))
print(dname, utils.tokgreen('micro F1: ' + str(f1)))
results += dname + '\t' + utils.tokgreen('micro F1: ' + str(f1)) + '\n'
"""
if cats is None:
f1, out_s = D.eval(org_dev_datasets[di][1],
predictions)
else:
f1, out_s = D.eval(org_dev_datasets[di][1],
predictions, cats)
print(dname, utils.tokgreen(out_s))
results = dname + '\t' + utils.tokgreen(
out_s) + '\n'
if dname == 'aida-A':
dev_f1 = f1
continue
if config['multi_instance']:
predictions = self.predict(data, n_best=n_best)
else: # including semisup
predictions = self.predict(data, n_best=1)
cats = None
if 'cat' in data[0][0]['raw']['conll_m']:
cats = []
for doc in data:
cats += [
m['raw']['conll_m']['cat'] for m in doc
]
if cats is None:
f1 = D.eval(org_dev_datasets[di][1], predictions)
else:
f1, tab = D.eval(org_dev_datasets[di][1],
predictions, cats)
pprint(tab)
print(dname, utils.tokgreen('micro F1: ' + str(f1)))
if dev_f1 >= best_f1 and dev_f1 >= f1_start_couting: # 0.82 (for weak supervised learning alone)
is_counting = True
not_better_count = 0
if is_counting:
if dev_f1 < best_f1:
not_better_count += 1
print('not dev f1 inc after', not_better_count)
else:
final_result_str = results
not_better_count = 0
best_f1 = dev_f1
if self.args.model_path is not None:
print('save model to', self.args.model_path)
self.model.save(self.args.model_path)
if not_better_count == self.args.n_not_inc:
print('dev f1 not inc after', not_better_count,
'... STOP')
stop = True
break
print('epoch', e, 'total loss', total_loss,
total_loss / len(train_dataset))
if stop:
print('**********************************************')
print('best results (f1 on aida-A):')
print(final_result_str)
break
def train(self, org_train_dataset, org_dev_datasets, config):
print('extracting training data')
train_dataset = self.get_data_items(org_train_dataset, predict=False)
print('#train docs', len(train_dataset))
dev_datasets = []
for dname, data in org_dev_datasets:
dev_datasets.append((dname, self.get_data_items(data,
predict=True)))
print(dname, '#dev docs', len(dev_datasets[-1][1]))
print('creating optimizer')
optimizer = optim.Adam(
[p for p in self.model.parameters() if p.requires_grad],
lr=config['lr'])
best_f1 = -1
not_better_count = 0
is_counting = False
eval_after_n_epochs = self.args.eval_after_n_epochs
for e in range(config['n_epochs']):
shuffle(train_dataset)
total_loss = 0
for dc, batch in enumerate(
train_dataset): # each document is a minibatch
self.model.train()
optimizer.zero_grad()
# convert data items to pytorch inputs
token_ids = [
m['context'][0] + m['context'][1]
if len(m['context'][0]) + len(m['context'][1]) > 0 else
[self.model.word_voca.unk_id] for m in batch
]
s_ltoken_ids = [m['snd_ctx'][0] for m in batch]
s_rtoken_ids = [m['snd_ctx'][1] for m in batch]
s_mtoken_ids = [m['snd_ment'] for m in batch]
entity_ids = Variable(
torch.LongTensor(
[m['selected_cands']['cands'] for m in batch]).cuda())
true_pos = Variable(
torch.LongTensor([
m['selected_cands']['true_pos'] for m in batch
]).cuda())
p_e_m = Variable(
torch.FloatTensor(
[m['selected_cands']['p_e_m'] for m in batch]).cuda())
entity_mask = Variable(
torch.FloatTensor(
[m['selected_cands']['mask'] for m in batch]).cuda())
token_ids, token_mask = utils.make_equal_len(
token_ids, self.model.word_voca.unk_id)
s_ltoken_ids, s_ltoken_mask = utils.make_equal_len(
s_ltoken_ids,
self.model.snd_word_voca.unk_id,
to_right=False)
s_rtoken_ids, s_rtoken_mask = utils.make_equal_len(
s_rtoken_ids, self.model.snd_word_voca.unk_id)
s_rtoken_ids = [l[::-1] for l in s_rtoken_ids]
s_rtoken_mask = [l[::-1] for l in s_rtoken_mask]
s_mtoken_ids, s_mtoken_mask = utils.make_equal_len(
s_mtoken_ids, self.model.snd_word_voca.unk_id)
token_ids = Variable(torch.LongTensor(token_ids).cuda())
token_mask = Variable(torch.FloatTensor(token_mask).cuda())
# too ugly but too lazy to fix it
self.model.s_ltoken_ids = Variable(
torch.LongTensor(s_ltoken_ids).cuda())
self.model.s_ltoken_mask = Variable(
torch.FloatTensor(s_ltoken_mask).cuda())
self.model.s_rtoken_ids = Variable(
torch.LongTensor(s_rtoken_ids).cuda())
self.model.s_rtoken_mask = Variable(
torch.FloatTensor(s_rtoken_mask).cuda())
self.model.s_mtoken_ids = Variable(
torch.LongTensor(s_mtoken_ids).cuda())
self.model.s_mtoken_mask = Variable(
torch.FloatTensor(s_mtoken_mask).cuda())
scores = self.model.forward(token_ids,
token_mask,
entity_ids,
entity_mask,
p_e_m,
gold=true_pos.view(-1, 1))
loss = self.model.loss(scores, true_pos)
loss.backward()
optimizer.step()
# self.model.regularize(max_norm=100)
loss = loss.cpu().data.numpy()
total_loss += loss
print('epoch',
e,
"%0.2f%%" % (dc / len(train_dataset) * 100),
loss,
end='\r')
print('epoch', e, 'total loss', total_loss,
total_loss / len(train_dataset))
if (e + 1) % eval_after_n_epochs == 0:
dev_f1 = 0
for di, (dname, data) in enumerate(dev_datasets):
predictions = self.predict(data)
f1 = D.eval(org_dev_datasets[di][1], predictions)
print(dname, utils.tokgreen('micro F1: ' + str(f1)))
if dname == 'aida-A':
dev_f1 = f1
if config[
'lr'] == 1e-4 and dev_f1 >= self.args.dev_f1_change_lr:
eval_after_n_epochs = 2
is_counting = True
best_f1 = dev_f1
not_better_count = 0
config['lr'] = 1e-5
print('change learning rate to', config['lr'])
if self.args.mulrel_type == 'rel-norm':
optimizer = optim.Adam([
p
for p in self.model.parameters() if p.requires_grad
],
lr=config['lr'])
elif self.args.mulrel_type == 'ment-norm':
for param_group in optimizer.param_groups:
param_group['lr'] = config['lr']
if is_counting:
if dev_f1 < best_f1:
not_better_count += 1
else:
not_better_count = 0
best_f1 = dev_f1
print('save model to', self.args.model_path)
self.model.save(self.args.model_path)
if not_better_count == self.args.n_not_inc:
break
self.model.print_weight_norm()
preranked_train = preranked_train[:min(args.n_docs,
len(preranked_train))]
org_dev_datasets = [(all_datasets[i][0], all_datasets[i][1])
for i in range(1, len(all_datasets))]
dev_datasets = [(all_datasets[i][0], all_datasets[i][2])
for i in range(1, len(all_datasets))]
dev_datasets = [(all_datasets[i][0], all_datasets[i][1],
all_datasets[2])
for i in range(1, len(all_datasets))]
for di, (dname, data, preranked) in enumerate(dev_datasets):
ranker.model._coh_ctx_vecs = []
predictions = ranker.predict(preranked)
print(
dname,
utils.tokgreen('micro F1: ' + str(D.eval(data, predictions))))
elif args.mode == 'ed':
with open(args.filelist, 'r') as flist:
for fname in flist:
fname = fname.strip()
print('load file from', fname)
conll_path = fname
cands_path = conll_path + '.csv'
data = D.CoNLLDataset.load_file(conll_path, cands_path,
person_path)
data = ranker.get_data_items(data, predict=True)
print('#docs', len(data))
continue
('reddit2020silver', conll.reddit2020silver),
('reddit2020g_s', conll.reddit2020g_s),
]
if args.mode == 'train':
print('training...')
config = {'lr': args.learning_rate, 'n_epochs': args.n_epochs}
pprint(config)
ranker.train(conll.train, dev_datasets, config)
elif args.mode == 'eval':
org_dev_datasets = dev_datasets # + [('aida-train', conll.train)]
dev_datasets = []
for dname, data in org_dev_datasets:
dev_datasets.append(
(dname, ranker.get_data_items(data, predict=True)))
print(dname, '#dev docs', len(dev_datasets[-1][1]))
vecs = ranker.model.rel_embs.cpu().data.numpy()
for di, (dname, data) in enumerate(dev_datasets):
ranker.model._coh_ctx_vecs = []
predictions = ranker.predict(data)
# **YD** change output of D.eval to include prec, rec and f1
"""
print(dname, utils.tokgreen('micro F1: ' + str(D.eval(org_dev_datasets[di][1], predictions))))
"""
f1, out_s = D.eval(data, predictions)
print(dname, utils.tokgreen(out_s))
pprint(config)
ranker.train(conll.train, dev_datasets, config)
elif args.mode == 'eval':
org_dev_datasets = dev_datasets # + [('aida-train', conll.train)]
dev_datasets = []
for dname, data in org_dev_datasets:
dev_datasets.append((dname, ranker.get_data_items(data, predict=True)))
print(dname, '#dev docs', len(dev_datasets[-1][1]))
if dname == 'test':
with open('test_data.pickle', 'wb') as f:
pickle.dump(ranker.get_data_items(data, predict=False), f, pickle.HIGHEST_PROTOCOL)
vecs = ranker.model.rel_embs.cpu().data.numpy()
for di, (dname, data) in enumerate(dev_datasets):
if di == 1:
break
ranker.model._coh_ctx_vecs = []
predictions = ranker.predict(data)
print(dname, utils.tokgreen('micro F1: ' + str(D.eval(org_dev_datasets[di][1], predictions))))
elif args.mode == 'test':
text = ''
with open(args.input_file, 'r', encoding='utf-8') as f:
for line in f.readlines():
text += line + '\n'
result = entity_linking_plain(text)
print(result)
