Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Deep span #19

Merged
merged 4 commits into from
Nov 15, 2021
Merged

Deep span #19

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
5cadad3
add SciBERT
syuoni Nov 12, 2021
c0f365c
fix bugs
syuoni Nov 15, 2021
4a2a1ac
fix bugs
syuoni Nov 15, 2021
ed12d10
update scripts
syuoni Nov 15, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions data/ace-luan2019naacl/ace-luan2019naacl-process.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,6 +35,8 @@
'head': spans.index((rel[0]-curr_start, rel[1]-curr_start+1)),
'tail': spans.index((rel[2]-curr_start, rel[3]-curr_start+1))} for rel in ex['relations'][k]]
new_data.append(new_ex)
curr_start += len(new_ex['tokens'])


with open(trg_fn, 'w') as f:
json.dump(new_data, f)
8 changes: 5 additions & 3 deletions scripts/exp_launcher.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -193,13 +193,15 @@ def call_command(command: str):
ck_label_emb_dim=[10, 25])
else:
sampler = OptionSampler(num_epochs=50,
lr=[1e-3, 2e-3],
finetune_lr=[1e-5, 2e-5],
# lr=[1e-3, 2e-3],
lr=numpy.logspace(-3.0, -2.5, num=100, base=10).tolist(), # 1e-3 ~ 3e-3
# finetune_lr=[5e-5, 1e-4],
finetune_lr=numpy.logspace(-4.5, -4.0, num=100, base=10).tolist(), # 3e-5 ~ 1e-4
batch_size=48,
ck_decoder='span_classification',
bert_drop_rate=0.2,
use_interm2=[False, True],
bert_arch=['BERT_base', 'RoBERTa_base'])
bert_arch=['BERT_base', 'RoBERTa_base', 'SciBERT'])

elif args.task == 'text2text':
COMMAND = " ".join([COMMAND, "@scripts/options/tf2text.opt"])
Expand Down
2 changes: 1 addition & 1 deletion scripts/exp_results_collector.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@


dict_re = re.compile("\{[^\{\}]+\}")
metircs_re = {'acc': re.compile("(?<=Accuracy: )\d+\.\d+(?=%)"),
metrics_re = {'acc': re.compile("(?<=Accuracy: )\d+\.\d+(?=%)"),
'micro_f1': re.compile("(?<=Micro F1-score: )\d+\.\d+(?=%)"),
'bleu4': re.compile("(?<=BLEU-4: )\d+\.\d+(?=%)")}

Expand Down
4 changes: 2 additions & 2 deletions scripts/joint_extraction.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -208,10 +208,10 @@ def save_callback(model):

logger.info("Evaluating on dev-set")
evaluate_joint_extraction(trainer, dev_set, has_attr=(args.attr_decoder!='None'), has_rel=(args.rel_decoder!='None'), eval_chunk_type_for_relation=True, batch_size=args.batch_size)
# evaluate_joint_extraction(trainer, dev_set, has_attr=(args.attr_decoder!='None'), has_rel=(args.rel_decoder!='None'), eval_chunk_type_for_relation=False, batch_size=args.batch_size)
evaluate_joint_extraction(trainer, dev_set, has_attr=(args.attr_decoder!='None'), has_rel=(args.rel_decoder!='None'), eval_chunk_type_for_relation=False, batch_size=args.batch_size)
logger.info("Evaluating on test-set")
evaluate_joint_extraction(trainer, test_set, has_attr=(args.attr_decoder!='None'), has_rel=(args.rel_decoder!='None'), eval_chunk_type_for_relation=True, batch_size=args.batch_size)
# evaluate_joint_extraction(trainer, test_set, has_attr=(args.attr_decoder!='None'), has_rel=(args.rel_decoder!='None'), eval_chunk_type_for_relation=False, batch_size=args.batch_size)
evaluate_joint_extraction(trainer, test_set, has_attr=(args.attr_decoder!='None'), has_rel=(args.rel_decoder!='None'), eval_chunk_type_for_relation=False, batch_size=args.batch_size)

logger.info(" ".join(sys.argv))
logger.info(pprint.pformat(args.__dict__))
Expand Down
64 changes: 50 additions & 14 deletions scripts/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -119,6 +119,7 @@ def parse_to_args(parser: argparse.ArgumentParser):
'ace2005': 'English',
'conll2004': 'English',
'SciERC': 'English',
'ace2005_rel': 'English',
'ResumeNER': 'Chinese',
'WeiboNER': 'Chinese',
'SIGHAN2006': 'Chinese',
Expand All @@ -136,6 +137,9 @@ def parse_to_args(parser: argparse.ArgumentParser):
'flickr8k': 'English',
'flickr30k': 'English',
'mscoco': 'English'}
dataset2language.update({f'ADE_cv{k}': 'English' for k in range(10)})
dataset2language.update({f'ace2004_rel_cv{k}': 'English' for k in range(5)})


def load_data(args: argparse.Namespace):
if args.dataset == 'conll2003':
Expand Down Expand Up @@ -194,22 +198,49 @@ def load_data(args: argparse.Namespace):
f"Corruption Retrieval F1-score: {ave_scores['micro']['f1']*100:2.3f}%")

elif args.dataset == 'conll2004':
json_io = JsonIO(text_key='tokens',
chunk_key='entities', chunk_type_key='type', chunk_start_key='start', chunk_end_key='end',
relation_key='relations', relation_type_key='type', relation_head_key='head', relation_tail_key='tail',
case_mode='None', number_mode='Zeros')
train_data = json_io.read("data/conll2004/conll04_train.json")
dev_data = json_io.read("data/conll2004/conll04_dev.json")
test_data = json_io.read("data/conll2004/conll04_test.json")
io = JsonIO(text_key='tokens',
chunk_key='entities', chunk_type_key='type', chunk_start_key='start', chunk_end_key='end',
relation_key='relations', relation_type_key='type', relation_head_key='head', relation_tail_key='tail',
case_mode='None', number_mode='Zeros')
train_data = io.read("data/conll2004/conll04_train.json")
dev_data = io.read("data/conll2004/conll04_dev.json")
test_data = io.read("data/conll2004/conll04_test.json")

elif args.dataset == 'SciERC':
json_io = JsonIO(text_key='tokens',
chunk_key='entities', chunk_type_key='type', chunk_start_key='start', chunk_end_key='end',
relation_key='relations', relation_type_key='type', relation_head_key='head', relation_tail_key='tail',
case_mode='None', number_mode='Zeros')
train_data = json_io.read("data/SciERC/scierc_train.json")
dev_data = json_io.read("data/SciERC/scierc_dev.json")
test_data = json_io.read("data/SciERC/scierc_test.json")
io = JsonIO(text_key='tokens',
chunk_key='entities', chunk_type_key='type', chunk_start_key='start', chunk_end_key='end',
relation_key='relations', relation_type_key='type', relation_head_key='head', relation_tail_key='tail',
case_mode='None', number_mode='Zeros')
train_data = io.read("data/SciERC/scierc_train.json")
dev_data = io.read("data/SciERC/scierc_dev.json")
test_data = io.read("data/SciERC/scierc_test.json")

elif args.dataset.startswith('ADE_cv'):
io = JsonIO(text_key='tokens',
chunk_key='entities', chunk_type_key='type', chunk_start_key='start', chunk_end_key='end',
relation_key='relations', relation_type_key='type', relation_head_key='head', relation_tail_key='tail',
case_mode='None', number_mode='Zeros')
k = int(args.dataset.replace('ADE_cv', ''))
train_data = io.read(f"data/ADE/ade_split_{k}_train.json")
dev_data = []
test_data = io.read(f"data/ADE/ade_split_{k}_test.json")
args.train_with_dev = True

elif args.dataset.startswith('ace2004_rel_cv'):
io = JsonIO(relation_key='relations', relation_type_key='type', relation_head_key='head', relation_tail_key='tail',
case_mode='None', number_mode='Zeros')
k = int(args.dataset.replace('ace2004_rel_cv', ''))
train_data = io.read(f"data/ace-luan2019naacl/ace04/cv{k}.train.json")
dev_data = []
test_data = io.read(f"data/ace-luan2019naacl/ace04/cv{k}.test.json")
args.train_with_dev = True

elif args.dataset == 'ace2005_rel':
io = JsonIO(relation_key='relations', relation_type_key='type', relation_head_key='head', relation_tail_key='tail',
case_mode='None', number_mode='Zeros')
train_data = io.read("data/ace-luan2019naacl/ace05/train.json")
dev_data = io.read("data/ace-luan2019naacl/ace05/dev.json")
test_data = io.read("data/ace-luan2019naacl/ace05/test.json")

elif args.dataset == 'ResumeNER':
conll_io = ConllIO(text_col_id=0, tag_col_id=1, scheme='BMES', encoding='utf-8', token_sep="", pad_token="")
Expand Down Expand Up @@ -389,6 +420,11 @@ def load_pretrained(pretrained_str, args: argparse.Namespace, cased=False):
return (transformers.BertModel.from_pretrained(PATH, hidden_dropout_prob=args.bert_drop_rate, attention_probs_dropout_prob=args.bert_drop_rate),
transformers.BertTokenizer.from_pretrained(PATH, model_max_length=512, do_lower_case=False))

elif pretrained_str.lower().startswith('scibert'):
PATH = "assets/transformers/allenai/scibert_scivocab_cased" if cased else "assets/transformers/allenai/scibert_scivocab_uncased"
return (transformers.BertModel.from_pretrained(PATH, hidden_dropout_prob=args.bert_drop_rate, attention_probs_dropout_prob=args.bert_drop_rate),
transformers.BertTokenizer.from_pretrained(PATH, model_max_length=512))

elif args.language.lower() == 'chinese':
if pretrained_str.lower().startswith('bert'):
PATH = "assets/transformers/hfl/chinese-bert-wwm-ext"
Expand Down