The Conversational Intelligence Challenge

NIPS 2017 Live Competition

Conversational Intelligence Challenge baseline solution

The solution is placed at GitHub, you could fork or download it here.

For now it is a Telegram bot, but it could easily be switched to our server by providing base_url parameter in telegram bot initialisation here. More information about parameters of telegram bot framework could be found here.

The solutaion is based on two papers:

We are using forked repo of Allen AI2 bi-att-flow: https://github.com/allenai/bi-att-flow

Requirements

Python packages will be installed by setup.sh script.

Cloning

It you’re using git version 1.6.5+, you could clone with dependencies:

git clone --recursive git@github.com:MIPTDeepLearningLab/ConvAI-baseline.git

If you’re using earlier versions of git, or if you have the repo downloaded without --recursive, you should clone dependency repos additionally:

git clone git@github.com:MIPTDeepLearningLab/ConvAI-baseline.git
cd ConvAI-baseline
git submodule update --init --recursive

Setup

Run setup.sh

Setup will download docker images, models and data files, so you have no need to download any of that by yourself.

Running

./bot.sh start|stop|stopall|restart|restartall|status

stopall is stopping all the services after the killing the bot, while stop just kills the bot only. The restarall is using stopall instead of simple stop.