Inspiration

Being a student of Machine Learning, I am constantly learning and it is very helpful for me to try out new opportunities and technologies. I am really passionate about open source contribution and have been contributing to it for the last 1 year. I have also been a facilitator for Facebook DevC Study groups where I arranged multiple workshops on Machine learning but did not have the time to implement something like this. but this time I fulfilled my dream of making such a tutorial so that even if a novice to Deep learning reads it, he is most likely to understand it and implement it on his first try. Also, Pytorch is so simple to implement so the learner wouldn't run into any issues related to that.

What it does

It implements a simple restaurant chatbot where you can chat with the AI Model as you would talk to a real person.

How I built it

It is a Chatbot that is implemented using a Feed-Forward PyTorch Neural Network with 3 layers with ReLU activation and Cross-Entropy Loss.

I built it with a combination of Python & Pytorch

Challenges I ran into

One of the biggest challenges that I came across was to implement this model in such a way that I can teach this complex topic to a learner who is new to DL

another technical issue was choosing the number of layers and adjusting its hyperparameters so it would not overfit

Accomplishments that I'm proud of

This was my first time implementing a Feed-Forward Neural network in Pytorch. I have implemented other models using Tensorflow but this was a first in Pytorch

What I learned

I am very much experienced with Natural Language Processing and Deep Learning but that was mostly through Tensorflow and Scikit learn. This was my first time implementing such a Deep Learning Model Using Pytorch and I had fun building it.

What's next for Building a Chatbot with Pytorch

I will be Extending this model to a full-fledged Deep Neural network that would perform Abstractive summarization of the user queries and respond accordingly. it will be trained on a much bigger dataset than this. and the most fun part of it will be that it will be in Pytorch.

Share this project:

Updates