zizhenwang / fbbc Goto Github PK

FBBC is abbreviation of Feature-Based BERT Classifier, and it is easy to adapt to other tasks with few changes.

1. Refactor DataProcessor to fit your task.
2. run following command as BERT does

   python simple_classifier.py --do_train=true --do_eval=true\
       --data_dir=YOUR_DATA_PATH --bert_config_file=CONFIG_PATH\
       --vocab_file=VOCAB_PATH --output_dir=OUTPUT_MODEL
   

   if you want to fine-tune BERT model, add arg --finetune=true
3. if you want to check a model is frozen or finetune, run

   python check_fintuned.py --raw_ckpt=BERT_MODEL_PATH --trained_ckpt=YOUR_MODEL_PATH
   

• simple_classifier.py

  In this code we can choose how to use pre-trained BERT model. Notice there is a finetune flag default by false which means this classifier uses BERT as feature-based model like ELMo. We can also set it to true just like the paper does.

  Another thing is we simplify the processors here.

• optimizeation.py

  We set trainable variables in this code. We use bert as keyword to filter BERT variables out, so you should name your variables without bert in it.