Code for the paper

Natural Language Does Not Emerge 'Naturally' in Multi-Agent Dialog
Satwik Kottur, José M. F. Moura, Stefan Lee, Dhruv Batra
Arxiv

This repository contains code to train, evaluate and visualize dialogs between conversational agents (Abot and QBot), which talk about instances in an abstract world.

If you find this code useful, consider citing our work:

@inproceedings{visdial,
  title = {{N}atural {L}anguage {D}oes {N}ot {E}merge '{N}aturally' in {M}ulti-{A}gent {D}ialog},
  author = {Satwik Kottur and Jos\'e M.F. Moura and Stefan Lee and Dhruv Batra},
  journal = {CoRR},
  volume = {abs/1706.08502},
  year = {2017}
}

All our code is implemented in Pytorch.

Additionally, our code also uses some famous python packages that can be installed as follows:

pip install json
pip install tqdm
pip install pickle
pip install json

• options.py - Read the options from the commandline
• dataloader.py - Create and handle data for toy instances
• chatbots.py - Conversational agents - Abot and Qbot
• learnChart.py - Obtain evolution of language chart from checkpoints
• html.py - Easy creation of html tables
• utilities.py - Helper functions
• train.py - Script to train conversational agents
• test.py - Script to test agents

Checkout run_me.sh to see how train our model.

Pretrained models and detailed documentation coming soon!

• Satwik Kottur

BSD