NO CODE DEEPLEARNING

Welcome to the NO CODE DEEPLEARNING project! This repository contains the source code for our hackathon project developed during FOSS HACK. Our goal was to create an intuitive, user-friendly interface that allows users to build neural networks by simply designing the blocks, adjusting hyperparameters, and downloading the resulting code.

The Problem We Are Solving

Deep learning model development often requires substantial programming knowledge and experience. Many individuals, especially those from non-technical backgrounds, find it challenging to experiment with neural networks due to the steep learning curve associated with coding and configuring these models. Our project aims to democratize deep learning by providing an intuitive, visual interface that simplifies the process of building and configuring neural networks.

Project Description

NO CODE DEEPLEARNING is an innovative platform designed to make deep learning accessible to everyone. Users can visually construct neural networks by dragging and dropping layers, adjusting hyperparameters for each layer, and ultimately downloading the generated PyTorch code. This approach eliminates the need for extensive programming knowledge and allows users to focus on the conceptual design of their models.

Note : During the hackathon our there was some issues with the main repository but at the end we mad a new repository, the old repository with the commits during the hackathon can be found below

https://github.com/NehalNetha/NoCodeDeepLearning

How It Works

1. Drag-and-Drop Interface: Users add and arrange layers to form a neural network.

2. Hyperparameter Configuration: Each layer's parameters, such as input size, number of neurons, and activation functions, can be customized.

3. Code Export: Once the network is designed, users can download the PyTorch code representing their model, ready for training and further development.

Project Timeline

Initial Stage (Conceptualization and Planning)

• Identified the problem of accessibility in deep learning model creation.

• Brainstormed potential solutions and settled on a visual, no-code interface.

• Defined the core features and technology stack (Next.js, Flask, PyTorch).

Development Stage

• Set up the development environment and repository structure.

• Developed the basic front-end interface with Next.js, including a drag-and-drop system for layers.

• Implemented the back-end with Flask, focusing on handling API requests and model configuration.

• Integrated PyTorch to generate and export model code based on the user's configuration.

• Conducted testing and debugging to ensure smooth functionality and user experience.

Current Stage

• Successfully developed the core functionality of the platform.

• Integrated the front-end and back-end to provide a seamless user experience.

• Implemented the code export feature, enabling users to download PyTorch code.

• Creating architectures in Feed Forward Neural Networks, Convolutional Neural Networks, and Recurrent Neural Networks.

• Getting the PyTorch file back after designing the architectures.

What Is Working

• The Building and Dropping blocks interface is intuitive and user-friendly.

• Hyperparameter configuration for each layer works seamlessly.

• Code export functionality generates accurate and ready-to-use PyTorch code.

• Creating architectures in Feed Forward Neural Networks, Convolutional Neural Networks, and Recurrent Neural Networks.

• Getting the PyTorch file back after designing the architectures.

More Features in the Future

1. Enabling to add more Architectures - Transformers, GANs, LSTMs.

2. Letting the User design the Data Transformations they want to Apply.

3. Better UI - Drag and Dropping.

4. More Robust Design of PyTorch Files.

Tech Stack

• Front-End: Next.js

• Back-End: Flask

• Deep Learning Framework: PyTorch

Build by -

• Khushin Vyas (https://github.com/khushinvyas)

• Nihal Netha (https://github.com/NehalNetha)