README.md

bertVersusAll

For Using the Traditional Embed-Encode-Attend-Predict Model

For Using the Bert Model (Minimal BERT)

1. Install: pip install -U bert-serving-server bert-serving-client

   • The server MUST be running on Python >= 3.5 with Tensorflow >= 1.10
   • Download a bert model: SMALL = https://storage.googleapis.com/bert_models/2018_10_18/cased_L-12_H-768_A-12.zip LARGE = https://storage.googleapis.com/bert_models/2018_10_18/cased_L-24_H-1024_A-16.zip
   • Unzip to a directory to point to later (Suggested inside venv folder)

2. Install other needed libraries and download data

   • pip install keras_metrics keras pandas numpy sklearn

3. Serve Model:

   • bert-serving-start -model_dir ~/bert/models/cased_L-12_H-768_A-12/ -num_worker=2 -max_seq_len=250 num_workers can be used to increase num_workers (Model is loaded n-times) max_seq_len specifies the largest entry size to the bert model (suggest 250 for concept work)