/Seq2Seq_chatbot_heroku

A Seq2Seq Attention chatbot deployed on Heroku

Primary LanguageJupyter Notebook

CHATBOT DEPLOYED ON HEROKU

This repo contains the code for training and Deployment of a Seq2Seq Attention based chatbot trained on Cornell Movie Dialogues.

Model Specifics

The Model comprises of 3 Encoder and 3 Decoder Layers connected by Bahdanau Attention. The Cornell Movie Dialogue corpus was fed into the model after preprocessing for rare words and unicode characters. The detailed preprocessing of the text and the training of the model code can be found in the Notebook attached.
I used pretrained Glove Embeddings for generating embeddings from the text before feeding it to model.

Running on Local machine

To run the model on local machine

  1. Open up a virtual environment (what is that?) and download all the packages in requirements.txt \ There might be some problems in installing tensorflow so, please install it manually using this

    pip install --upgrade pip

    pip install --upgrade tensorflow

    If there is still a problem, refer this

  2. Next, get the model_data folder here and put it in the same directory as app.py

  3. Now From the root directory i.e. The directory containing app.py run python app.py.
    This should start a local host server in your machine and you can open it up in the browser of your choice.

Link

The deployed Chatbot can be found at chat_link Please give it some time to load, if it takes some time.

Suggestions

If you have any ideas about speeding up loading time or any suggestions regarding code, I will be grateful to hear you.