CopyNet implementation with TensorFlow 2

• Incorporating Copying Mechanism in Sequence-to-Sequence Learning
• Uses TensorFlow 2.0 and above APIs with tf.keras too
• Adapted from AllenNLP's PyTorch implementation, their blog referenced below was very helpful to understand the math from an implementation perspective

Environment to run examples

Setup

• Copy sample.env to .env and enter appropriate values for the variables
• A brief description of each is provided as a comment in that file
• Post that run,

  ./setup-env.sh [--no-docker]

• Uses env file to configure project environment
• Builds required docker images (if you don't wanna use Docker then pass --no-docker option to the setup-env.sh script)
• Makes a python environment and installes required packages in it
• Prepares an lock.env file. Do not edit/ delete it

Rebuilding environment

• You may change environment config in the process of development
• This includes adding a new python package to requirements.txt
• After changing run,

  ./setup-env.sh [--no-docker]
  

• If you do not want Docker, then pass --no-docker option similar to before

Start environment

• At the end of setup script you will be shown the commands to start the environments
• They are,

  ./start-env.sh nb    # For Dockerized jupyter server
  ./start-env.sh bash  # For Dockerized bash

• It is not necessary to use the start-env.sh script for virtualenv, the regular source command to activate it is enough

Note on Dockerized environment

• The dockerized environment is specifically helpful and recommended when using GPU
• It takes care of many nuances involved in setting up CUDA. Your host machine should just have correct NVIDIA drivers and nothing else
• It is recommended to run the examples in this environment to ensure all correct dependencies are met

Run examples

• Instructions to run an example are detailed in its own folders respectively

References

• Incorporating Copying Mechanism in Sequence-to-Sequence Learning: (paper)
• AllenNLP implementation: (blog) (code)
• BLEU score metric: (code)