NIPS Challenge Docker container

This repository contains the Dockerfile and setup code to run chat bot in docker instance.

Live Deployment

Telegram bot @conv_test. Make sure to have an username registered in Telegram, and then start conversation with \begin.

File description

• bot.py : Main entry point of the chat bot, message selection logic can be implemented here.
• models/ : Folder where model code is stored
• data/ : Folder where data files are stored
• config.py : Configuration script, which has the location of data files as well as bot tokens. Replace the bot token with your ones to test.
• models/wrappper.py - Wrapper function which calls the models. Must implement get_response def.
• models/setup - shell script to download the models
• data/setup - shell script to download the data files and saved model files
• model_selection.py - Selection logic for best answer

Running Docker

• After installing docker, build the image from this directory using the following command: docker build -t convai .
• Docker will create a virtual container with all the dependencies needed.
• Docker will autostart the bot whenever the container is run: docker run convai

Adding your own models

• In models/setup, add the repository of your model (should be a public repository for now) to clone.
• In data/setup, add the data location to download your saved model data
• Change the config.py with the endpoint of the data
• Create a wrapper in models/wrapper.py for your model
• Modify the model_selection.py to call your model.

Bugs

Feel free to open an issue or submit a PR.

Authors

Nips ConvAI Challenge McGill RLLDialog Team