Tensorflow Chatbot

Tensorflow Chatbot Demo by @Sirajology on Youtube

Overview

This is the full code for 'How to Make an Amazing Tensorflow Chatbot Easily' by @Sirajology on Youtube. In this demo code, we implement Tensorflows Sequence to Sequence model to train a chatbot on the Cornell Movie Dialogue dataset. After training for a few hours, the bot is able to hold a fun conversation.

Dependencies

• numpy
• scipy
• six
• tensorflow (https://www.tensorflow.org/versions/r0.12/get_started/os_setup.html)

Use pip to install any missing dependencies

Usage

Create venv & install dependencies:

Using viritualenv here to be compatable with python2, install with "pip install virtualenv"

# create venv
python -m virtualenv .venv

# enter venv (assuming macos/linux)
source .venv/bin/activate

# install requirements
pip install -r requirements.txt

Training

0. Create directories

   mkdir working_dir
   mkdir data
   

1. Download the Cornell Movie Dialogue dataset and place the unzipped content in the data/ directory.

   cd data
   wget http://www.mpi-sws.org/~cristian/data/cornell_movie_dialogs_corpus.zip
   unzip cornell_movie_dialogs_corpus.zip
   
   # move to `data/` root
   mv "corenell movie-dialogs corpus"/* .
   

2. Prepare data for training, in the data/ directory, run the prepare_data.py script

   # move to project root
   cd ..
   python prepare_data.py
   

3. To train the bot, edit the seq2seq.ini file so that mode is set to train like so

mode = train

4. Start training, by running the code like so:

   python execute.py

There is no mechanism to stop training, you will need to 'ctrl-c' to stop training after a period of time.

Test

1. To test the bot during or after training, edit the seq2seq.ini file so that mode is set to test like so

   mode = test

2. To test run the code like so:

   python execute.py
   
   >> Mode : test
   
   Reading model parameters from working_dir/seq2seq.ckpt-10200
   >    
   

3. Confrim...

   • What does "Test do?"
   • How to use it?

Challenge

The challenge for this video is write an entirely different script using TF Learn to generate Lord of the Ring style sentences. Check out this very similar example, it uses TF Learn to generate Shakespeare-style sentences. Train your model on Lord of the rings text to do something similar! And play around with the hyperparameters to get a more accurate result. Post your GitHub link in the video comments and I'll judge it!

Due date: December 8th

Also see this issue, some people have found this discussion helpful llSourcell#3

Credits

Credit for the vast majority of code here goes to suriyadeepan. I've merely created a wrapper to get people started.