Google smart reply 2017 implementation in tensorflow
Clone or download
Pull request Compare This branch is even with clsk:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
dataset
models
.gitignore
LICENSE
README.md
main_dual_encoder_dense.py
requirements.txt

README.md

google-smart-reply-2017

Google smart reply paper (2017) implementation in tensorflow

Getting started

  1. Get Ubuntu corpus dataset for testing from here
wget https://s3.amazonaws.com/ngv-public/data.zip -O data.zip
  1. Unzip and move data files wherever you want.
unzip data.zip -d .
  1. Install conda environment
conda create -n sr python=3.6
pip install -r requirements.txt

source activate sr
  1. Update the path variables with links to the data and where you want to save model output
# main_dual_encoder_dense.py
# path params
parser.add_argument('--root_dir', default='')
parser.add_argument('--dataset_train_path', default='')
parser.add_argument('--dataset_test_path', default='')
parser.add_argument('--dataset_val_path', default='')
parser.add_argument('--vocab_path', default='')

parser.add_argument('--model_save_dir', default='')
parser.add_argument('--test_tube_dir', default='')
  1. Start training
python main_dual_encoder_dense.py