Seq2seq Attention Bot

It is a chatbot with seq2seq neural network with basic attention mechanism, completely implemented in Python using Tensorflow 2.0 and keras package. Here we use Cornell Movie Corpus Dataset!

The follwoing steps are needed to be performed to run the chatbot.

1. Choose any version to work with.
2. The data is first needed to be preprocessed using preprocess.py. It then creates two languages ( one for questions and one for answers) and saves them.
3. The file Chatbot_class.py contains the python implementation of our seq2seq model using basic RNN and GRU Cell.
4. The file training.py needs to be executed to run the training over the dataset. This training step differs between v_1 and v_2. It creates two .pickle file called encoder and decoder which contians the trained parameters. Note: This step may take a while. It is recommended to use a GPU to perform the training.
5. Finally bot_app.py can be executed to run the chatbot from console.

In this model, we use Cornell-Corpus Movie dialogue dataset for training the chatbot.

1. Extract the .zip file contaiing the Movie Dialogue dataset.

2. Run preprocess.py create the language specific files and the tokenizers, where we store all the language information.

3. Run Training.py to train the dataset

4. If you run it for the first time, set first_time = True, it will save the encoder and decoder after training

5. If you want to train it more, set first_time = False, it will load the poreviously saved encoder and decoder and start training from it.

6. You need to run bot_app.py after training to run the chatbot in an interactive mode from console.

7. bot_app.py implements 3 different sampling schemes in order to generate specific replies to the query.