0) python 3.6
1) numpy 1.18.1
2) sklearn 0.22.1
3) matplotlib 3.1.3
4) torch 1.4.0
5) torchvision 0.5.0
6) jieba 0.42.1
7) transformers 2.4.1
8) pytorch-crf 0.7.2
./data/d1/: the dir of dataset
./data/s1/: the dir of solution
./data/vocab.txt: the pretrained tokenize dict used by bert
./code: the dir of code
./readme.md
./run.sh
./run_local.py
./requirements.txt
bash run.sh
To change the task mode, you can choose from train, eval, predict in the file run.sh:
--mode='train'
To change the dataset for training, you need to change the file run.sh:
--dataset_dir='./data/d5'
--answer_dir='./data/s5'
--task='xxx'
To change to different model for training, you need to change the file ./code/train.py(main), such as
# mymodel = BLSTM_CRF(model_params, show_param=True)
# mymodel = BERT_LSTM_CRF(model_params, show_param=True)
# mymodel = BERT_MLP(model_params, show_param=True)
mymodel = BERT_NER(model_params, show_param=True)
# mymodel = BERT_CRF(model_params, show_param=True)
To change the training parameters such as learning_rate, batch_size, training_epoch, you need to change the file ./code/train.py(_train)
train_param = {
'EPOCH': 15,
'batch_size': 64,
'learning_rate_bert': 5e-5,
'learning_rate_upper': 5e-3,
'bert_finetune': True,
'visualize_length': 20,
'isshuffle': True,
'result_dir': args.result_dir,
'model_name':'model_test.p'
}
it will generate a fold at the root dir such as ./result/ and a log file ./log_info.txt, the ./result/ file containing:
joint_embedding_info_dict.pkl #preprocess dict info
model_test.p # trained model result
test_data_mat_dict.pkl #preprocessed test dataset
train_data_mat_dict.pkl #preprocessed train dataset
loss_train.txt
score_train.txt
loss.png
score.png