CommonParams(
dataset_name="duee1.0",
experiment_name="commodity_ner",
train_file=train_text_file,
eval_file=train_bio_file,
#test_file=test_file,
learning_rate=2e-5,
train_max_seq_length=512,
eval_max_seq_length=512,
per_gpu_train_batch_size=4,
per_gpu_eval_batch_size=4,
per_gpu_predict_batch_size=4,
#per_gpu_train_batch_size=16,
#per_gpu_eval_batch_size=16,
#per_gpu_predict_batch_size=16,
seg_len=510,
seg_backoff=128,
num_train_epochs=10,
fold=0,
#num_augements=3,
enable_kd=False,
enable_sda=False,
sda_teachers=2,
loss_type="CrossEntropyLoss",
# loss_type='FocalLoss',
focalloss_gamma=2.0,
model_type="bert",
model_path=
# "/opt/share/pretrained/pytorch/hfl/chinese-electra-large-discriminator",
r"E:\ai_contest\ccks\mytheta\myRepo\theta-master-0826\theta\examples\LIC2021\model_rbt3",
# "/opt/share/pretrained/pytorch/bert-base-chinese",
fp16=False,
best_index="f1",
random_type="np"),
CommonParams(
dataset_name="xfyun",
experiment_name="xfyun_ner",
train_file=train_file,
eval_file=None,
test_file=test_file, #test_spo_file,
learning_rate=2e-5,
train_max_seq_length=256,
eval_max_seq_length=256,
per_gpu_train_batch_size=16,
per_gpu_eval_batch_size=16,
per_gpu_predict_batch_size=16,
#per_gpu_train_batch_size=16,
#per_gpu_eval_batch_size=16,
#per_gpu_predict_batch_size=16,
seg_len=510,
seg_backoff=128,
num_train_epochs=10,
fold=0,
num_augements=0,
enable_kd=False,
enable_sda=False,
sda_teachers=2,
loss_type="CrossEntropyLoss",
# loss_type='FocalLoss',
focalloss_gamma=2.0,
model_type="bert",
model_path=
# "/opt/share/pretrained/pytorch/hfl/chinese-electra-large-discriminator",
#r"/opt/kelvin/python/knowledge_graph/theta/theta-master-0826/theta/examples/LIC2021/model_rbt3",
r"/kaggle/working",
fp16=False,
best_index="f1",
random_type="np"),
CommonParams(
dataset_name="event_element",
experiment_name="ccks2020_event_element",
train_file="data/event_element_train_data_label.txt",
eval_file="data/event_element_train_data_label.txt",
test_file="data/event_element_dev_data.txt",
learning_rate=2e-5,
train_max_seq_length=512,
eval_max_seq_length=512,
per_gpu_train_batch_size=4,
per_gpu_eval_batch_size=4,
per_gpu_predict_batch_size=4,
seg_len=510,
seg_backoff=128,
num_train_epochs=10,
fold=0,
num_augements=3,
enable_kd=False,
enable_sda=False,
sda_teachers=2,
loss_type="CrossEntropyLoss",
# loss_type='FocalLoss',
focalloss_gamma=2.0,
model_type="bert",
model_path=
# "/opt/share/pretrained/pytorch/hfl/chinese-electra-large-discriminator",
"/kaggle/working",
# "/opt/share/pretrained/pytorch/bert-base-chinese",
fp16=True,
best_index="f1",
random_type="np"),
experiment_params = GlueAppParams(
CommonParams(
dataset_name="o2o_reviews",
experiment_name="o2o_reviews",
train_file="data/train.csv",
eval_file="data/train.csv",
test_file="data/test_new.csv",
learning_rate=1e-5,
train_max_seq_length=256,
eval_max_seq_length=256,
per_gpu_train_batch_size=16,
per_gpu_eval_batch_size=16,
per_gpu_predict_batch_size=16,
seg_len=0,
seg_backoff=0,
num_train_epochs=10,
fold=0,
num_augements=0,
enable_kd=False,
enable_sda=False,
sda_teachers=1,
loss_type="CrossEntropyLoss",
model_type="bert",
model_path="/opt/share/pretrained/pytorch/bert-base-chinese",
train_rate=0.9,
fp16=False,
best_index='f1',
), GlueParams(glue_labels=glue_labels, ))
experiment_params.debug()
CommonParams(
dataset_name="event_classification",
experiment_name="ccks2020_entity_element",
train_file="data/event_element_train_data_label.txt",
eval_file="data/event_element_train_data_label.txt",
test_file="data/event_element_dev_data.txt",
learning_rate=1e-5,
train_max_seq_length=512,
eval_max_seq_length=512,
per_gpu_train_batch_size=4,
per_gpu_eval_batch_size=4,
per_gpu_predict_batch_size=4,
seg_len=0,
seg_backoff=0,
num_train_epochs=10,
fold=0,
num_augements=0,
enable_kd=False,
enable_sda=False,
sda_teachers=3,
sda_stategy="clone_models",
sda_empty_first=True,
loss_type="CrossEntropyLoss",
# loss_type="FocalLoss",
focalloss_gamma=1.5,
model_type="bert",
model_path=
"/opt/share/pretrained/pytorch/roberta-wwm-large-ext-chinese",
train_rate=0.9,
fp16=True,
best_index='f1',
),
experiment_params = NerAppParams(
CommonParams(
dataset_name="medical_entity",
experiment_name="ccks2020_medical_entity",
train_file='data/rawdata/ccks2020_2_task1_train/task1_train.txt',
eval_file='data/rawdata/ccks2020_2_task1_train/task1_train.txt',
test_file='data/rawdata/ccks2_task1_val/task1_no_val_utf8.txt',
learning_rate=2e-5,
train_max_seq_length=512,
eval_max_seq_length=512,
per_gpu_train_batch_size=4,
per_gpu_eval_batch_size=4,
per_gpu_predict_batch_size=4,
seg_len=510,
seg_backoff=128,
num_train_epochs=10,
fold=1,
num_augements=2,
enable_kd=True,
enable_sda=False,
sda_teachers=2,
loss_type="CrossEntropyLoss",
model_type="bert",
model_path=
# "/opt/share/pretrained/pytorch/hfl/chinese-electra-large-discriminator",
"/opt/share/pretrained/pytorch/roberta-wwm-large-ext-chinese",
fp16=True,
best_index="f1",
seed=6636,
random_type=None),
NerParams(ner_labels=ner_labels, ner_type='span', no_crf_loss=False))
from theta.modeling import Params, CommonParams, NerParams, NerAppParams, log_global_params
experiment_params = NerAppParams(
CommonParams(
dataset_name="medical_event",
experiment_name="ccks2020_medical_event",
tracking_uri="http://tracking.mlflow:5000",
train_file='data/task2_train_reformat.tsv',
eval_file='data/task2_train_reformat.tsv',
test_file='data/task2_no_val.tsv',
learning_rate=2e-5,
train_max_seq_length=512,
eval_max_seq_length=512,
per_gpu_train_batch_size=4,
per_gpu_eval_batch_size=4,
per_gpu_predict_batch_size=4,
seg_len=510,
seg_backoff=128,
num_train_epochs=10,
fold=3,
num_augements=2,
enable_kd=True,
loss_type="CrossEntropyLoss",
model_type="bert",
model_path=
"/opt/share/pretrained/pytorch/roberta-wwm-large-ext-chinese",
fp16=True,
random_type='np',
), NerParams(ner_labels=ner_labels, ner_type='crf', no_crf_loss=True))
experiment_params.debug()
CommonParams(
dataset_name="entity_typing",
experiment_name="ccks2020_entity_typing",
train_file="data/rawdata/ccks_7_1_competition_data/entity_type.txt",
eval_file="data/rawdata/ccks_7_1_competition_data/entity_type.txt",
test_file=
"data/rawdata/ccks_7_1_competition_data/entity_validation.txt",
learning_rate=1e-5,
train_max_seq_length=32,
eval_max_seq_length=32,
per_gpu_train_batch_size=96,
per_gpu_eval_batch_size=96,
per_gpu_predict_batch_size=96,
seg_len=0,
seg_backoff=0,
num_train_epochs=10,
fold=8,
num_augements=0,
enable_kd=False,
enable_sda=False,
sda_teachers=3,
sda_stategy="clone_models",
sda_empty_first=True,
# loss_type="CrossEntropyLoss",
loss_type="FocalLoss",
focalloss_gamma=2.0,
model_type="bert",
model_path=
"/opt/share/pretrained/pytorch/roberta-wwm-large-ext-chinese",
train_rate=0.9,
fp16=False,
best_index='f1',
),