AutoML

Logo
HomePage
Choices
Model Training Page

Inspiration

The inspiration of this project came from one of the sponsors in HTN (Co:here). Their goal is to make AI/ML accessible to devs, which gave me the idea, that I can build a platform, where people who do not even know how to code can build their own Machine Learning models. Coding is a great skill to have, but we need to ensure that it doesn't become a necessity to survive. There are a lot of people who prefer to work with the UI and cannot understand code. As developers, it is our duty to cater to this audience as well. This is my inspiration and goal through this project.

What it does

The project works by taking in the necessary details of the Machine Learning model that are required by the function. Then it works in the backend to dynamically generate code and build the model. It is even able to decide whether to convert data in the dataset to vectors or not, based on race conditions, and ensure that the model doesn't fail. It then returns the required metric to the user for them to check it out.

How I built it

I first built a Flask backend that took in information regarding the model using JSON. Then I built a service to parse and evaluate the necessary conditions for the Scikit Learn models, and then train and predict with it. After ensuring that my backend was working properly, I moved to the front-end where I spent a lot of my time, building a clean UI/UX design so that the users can have the best and the most comfortable experience while using my application.

Challenges I ran into

One of the key challenges of this project is to generate code dynamically at run-time upon user input. This requirement is a very hefty one as I had to ensure that the inputs won't break the code. I read through the documentation of Scikit Learn and worked with it, while building the web app.

Accomplishments that I'm proud of

I was able to full-fledged working application on my own, building the entire frontend and backend from scratch. The application is able to take in the features of the model and the dataset, and display the results of the trainings. This allows the user to tweak their model and check results everytime to see what works best for them. I'm especially proud of being able to generate and run code dynamically based on user input.

What I learned

This project, more than anything, was a challenge to myself, to see how far I had come from my last HackTheNorth experience. I wanted to get the full experience of building every part of the project, and that's why I worked solo. This gave me experience of all the aspects of building a software from scratch in limited amount of time, allowing me to grasp the bigger picture.

What's next for AutoML

My very first step will be to work on integrating TensorFlow to this project. My initial goal was to have a visual representation of Neural Network layers for users to drag and drop. Due to time and technical constraints, I couldn't fully idealize my goal. So this is the first thing I am going to work with. After this, I'll probably work with authentication so that people can work on their projects and store their progresses.