AI-Negotiation

Uses python 3.6

Data

Source of the dataset from Facebook AI Research: https://code.facebook.com/posts/1686672014972296/deal-or-no-deal-training-ai-bots-to-negotiate/

Virtual Environment

To create virtual environment:

sudo pip install virtualenv        # If virtualenv has not already been installed
virtualenv -p python3 .env         # To create a python3 virtual environment
source .env/bin/activate           # Activate the virtual environment
pip install -r requirements.txt    # Install all dependencies (only need to run once)
......
deactivate                         # Exit virtual environment

Files

Use readData.py to process the data from .txt files to a highly-formatted list of dictionaries for each example. The formatted data is saved in .json files

All of the models are defined in rnn.py

Train the negotiation output classifier using trainOutputClassification.py

Train the encoder and decoder RNNs to generate natural langauge responses in negotiation using trainGenerative.py

Converse with negotiation bots, either one operating by handcrafted rules or the RNN encoder/decoder bot using converse.py