For setup pytorch-BERT please follow huggingface-README.md. once you setup your environment , clone this repo and follow below execution steps.
This Repo is having code for downward NLP applications like Named Entity Recognition(NER),Intent classification using multi label classification,...etc.
this repo uses code from https://github.com/huggingface/pytorch-pretrained-BERT
pytorch_pretrained_bert version : 0.5.1 torch version : 1.0.0 numpy version : 1.15.4
Future work : sentiment analysis, QA retrival, text entailment, text similarity, text summarization..etc.
- 1st unzip files inside data directory of each sub-directory if zip file there else collect file but follow conll standard like inside data directory. remember input data structure different for each NLP application.
python examples/nerTest_usingBert.py --data_dir data/conll2003/ --bert_model bert-base-uncased --do_lower_case --do_train --do_eval --do_test --do_pred --task_name NER --output_dir custom_models/ner_output_conll2003
python examples/nerTest_usingBert.py --data_dir data/ncbi-disease/ --bert_model bert-base-uncased --do_lower_case --do_train --do_eval --do_test --do_pred --task_name NER --output_dir custom_models/ner_output_disease
python examples/intentTest_usingBert.py --data_dir data/toxic_cmt/ --bert_model bert-base-uncased --do_lower_case --do_train --do_eval --do_pred --task_name intent_multilabel --output_dir custom_models/intent_output_toxic
python examples/nerTest_usingBert.py --data_dir data/conll2003/ --bert_model bert-base-uncased --do_lower_case --do_train --do_eval --do_test --do_pred --task_name NER --output_dir custom_models/model-directory-name
python examples/nerTest_usingBert.py --data_dir data/ncbi-disease/ --bert_model bert-base-uncased --do_lower_case --do_train --do_eval --do_test --do_pred --task_name NER --output_dir custom_models/model-directory-name
python examples/intentTest_usingBert.py --data_dir data/toxic_cmt/ --bert_model bert-base-uncased --do_lower_case --do_pred --task_name intent_multilabel --output_dir custom_models/model-directory-name