-
Notifications
You must be signed in to change notification settings - Fork 2
Home
cha-no edited this page May 3, 2021
·
2 revisions
- data_dir, model_dir 등등 공통적인 argument는 제외했습니다
- baseline과 다른 부분은 bold체와 기울기로 구분했습니다
- 점수기준은 제출했을때 joint accuracy 기준입니다
- 공통 HyperParameter와 architecture의 HyperParameter를 나누었습니다
- 점수 : 0.6213
- epoch이 13일때 성능입니다
| 'random_seed' | 'architecture' | 'group_decay' | 'weight_decay' | 'max_seq_length' | 'max_label_length' | 'train_batch_size' | 'eval_batch_size' | 'learning_rate' | 'adam_epsilon' | 'max_grad_norm' | 'num_train_epochs' | 'warmup_ratio' | 'model_name_or_path' |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2020 | 'SUMBT' | True | 0.01 | 128 | 12 | 4 | 8 | 5e-5 | 1e-8 | 1.0 | 15 | 0.1 | 'dsksd/bert-ko-small-minimal' |
| 'hidden_dim' | 'num_rnn_layers' | 'zero_init_rnn' | 'attn_head' | 'fix_utterance_encoder' | 'task_name' | 'distance_metric' |
|---|---|---|---|---|---|---|
| 512 | 1 | False | 4 | False | 'sumbtgru' | 'euclidean' |
- 점수 : 0.6149
- epoch이 11일때 성능입니다
| 'random_seed' | 'architecture' | 'group_decay' | 'weight_decay' | 'max_seq_length' | 'max_label_length' | 'train_batch_size' | 'eval_batch_size' | 'learning_rate' | 'adam_epsilon' | 'max_grad_norm' | 'num_train_epochs' | 'warmup_ratio' | 'model_name_or_path' |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2020 | 'SUMBT' | True | 0.01 | 128 | 12 | 4 | 8 | 5e-5 | 1e-8 | 1.0 | 15 | 0.1 | 'dsksd/bert-ko-small-minimal' |
| 'hidden_dim' | 'num_rnn_layers' | 'zero_init_rnn' | 'attn_head' | 'fix_utterance_encoder' | 'task_name' | 'distance_metric' |
|---|---|---|---|---|---|---|
| 512 | 2 | False | 4 | False | 'sumbtgru' | 'euclidean' |
- 점수 : 0.6142
- epoch이 12일때 성능입니다
| 'random_seed' | 'architecture' | 'group_decay' | 'weight_decay' | 'max_seq_length' | 'max_label_length' | 'train_batch_size' | 'eval_batch_size' | 'learning_rate' | 'adam_epsilon' | 'max_grad_norm' | 'num_train_epochs' | 'warmup_ratio' | 'model_name_or_path' |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2020 | 'SUMBT' | True | 0.01 | 96 | 12 | 8 | 8 | 5e-5 | 1e-8 | 1.0 | 15 | 0.1 | 'dsksd/bert-ko-small-minimal' |
| 'hidden_dim' | 'num_rnn_layers' | 'zero_init_rnn' | 'attn_head' | 'fix_utterance_encoder' | 'task_name' | 'distance_metric' |
|---|---|---|---|---|---|---|
| 512 | 1 | False | 4 | False | 'sumbtgru' | 'euclidean' |
- 점수 : 0.6128
- epoch이 11일때 성능입니다
| 'random_seed' | 'architecture' | 'group_decay' | 'weight_decay' | 'max_seq_length' | 'max_label_length' | 'train_batch_size' | 'eval_batch_size' | 'learning_rate' | 'adam_epsilon' | 'max_grad_norm' | 'num_train_epochs' | 'warmup_ratio' | 'model_name_or_path' |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2020 | 'SUMBT' | True | 0.01 | 128 | 16 | 4 | 8 | 5e-5 | 1e-8 | 1.0 | 15 | 0.1 | 'dsksd/bert-ko-small-minimal' |
| 'hidden_dim' | 'num_rnn_layers' | 'zero_init_rnn' | 'attn_head' | 'fix_utterance_encoder' | 'task_name' | 'distance_metric' |
|---|---|---|---|---|---|---|
| 512 | 1 | False | 4 | False | 'sumbtgru' | 'euclidean' |
- 점수 : 0.6066
- epoch이 15일때 성능입니다
| 'random_seed' | 'architecture' | 'group_decay' | 'weight_decay' | 'max_seq_length' | 'max_label_length' | 'train_batch_size' | 'eval_batch_size' | 'learning_rate' | 'adam_epsilon' | 'max_grad_norm' | 'num_train_epochs' | 'warmup_ratio' | 'model_name_or_path' |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2020 | 'SUMBT' | True | 0.01 | 96 | 12 | 8 | 8 | 5e-5 | 1e-8 | 1.0 | 15 | 0.1 | 'dsksd/bert-ko-small-minimal' |
| 'hidden_dim' | 'num_rnn_layers' | 'zero_init_rnn' | 'attn_head' | 'fix_utterance_encoder' | 'task_name' | 'distance_metric' |
|---|---|---|---|---|---|---|
| 300 | 1 | False | 4 | False | 'sumbtgru' | 'euclidean' |
- 점수 : 0.6022
- epoch이 18일때 성능입니다
| 'random_seed' | 'architecture' | 'group_decay' | 'weight_decay' | 'max_seq_length' | 'max_label_length' | 'train_batch_size' | 'eval_batch_size' | 'learning_rate' | 'adam_epsilon' | 'max_grad_norm' | 'num_train_epochs' | 'warmup_ratio' | 'model_name_or_path' |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2020 | 'SUMBT' | True | 0.01 | 96 | 12 | 8 | 8 | 5e-5 | 1e-8 | 1.0 | 20 | 0.1 | 'dsksd/bert-ko-small-minimal' |
| 'hidden_dim' | 'num_rnn_layers' | 'zero_init_rnn' | 'attn_head' | 'fix_utterance_encoder' | 'task_name' | 'distance_metric' |
|---|---|---|---|---|---|---|
| 300 | 2 | False | 4 | False | 'sumbtgru' | 'euclidean' |