This repository contains the source code for dl.web.js, a platform for federated learning on the browser.

• Web app: https://dlonweb.netlify.app/
• Documentation and project details : https://shubhvjain.github.io/dlonweb/

• Simply go to https://dlonweb.netlify.app
• Upload your input, select a model and get the output - as simple as that!
• The inputs you upload stay on your browser

• You can easily do that as well or just run it on your local machine
• We provide a docker file for easy build and deployment of all app components

Due to resource constraints on the browser, we also provide a standalone backend service accessible via HTTP. It enables additional functionalities like converting Keras models for browser compatible inference

• Step 1: Ensure that Docker is installed on your local machine. You can find installation instructions here.

• Step 2: Clone the GitHub repository. This will download the latest code for the backend server from GitHub.

  • Command: git clone -b release https://github.com/shubhvjain/dlonweb.git

• Step 3: Open a terminal and navigate to the project's root directory.

• Step 4: Run the Docker Compose command.

  • The docker-compose.yml file in the root directory can be used to start both the frontend and backend services.
  • To run only the backend, use: docker compose up backend
  • To run both the frontend and backend, use: docker compose up

• Step 5: The backend server runs by default on port 3000. Make sure to specify this URL in your frontend configuration.

The project is divided into 3 main sub folders.

• The shared JS npm package with core abstract base classes for loading data , training the model and performing an inference task.
• It is used as a local dependency in the frontend and backend packages and enables consistent behavior across the browser and server environment.
• It is integrated using dependency injection allowing frontend and backend to plug in platform specific dependencies to avoid duplicate logic. For instance, using @tensorflow/tfjs package in the frontend and @tensorflow/tfjs-node in the backend to use Tensorflow.
• See the README file for more details

• Web-based user interface.
• The main app.
• Build using the Svelte framework.
• Styling using Bootstrap 5
• See the README file for more details

• Node.js backend with Python (Poetry) component.
• This serves and an additional,optional standalone service that can increase the power of the frontend. Needs to be run on a local port and is accessed via http from the frontend.
• See the README file for more details

• install.sh : Script to install all the whole project on a local computer. See the installation section for more details
• docs : this folder contains documentation files along with some demo input and output data
• backend.Dockerfile : the docker file for backend
• compose.yml : The docker compose file for the project
• frontend.Dockerfile : the docker file for frontend
• frontend.nginx.conf : the nginx config file used in the docker container to serve frontend
• netlify.toml : the configuration file for deploying frontend on netlify

Step to get the system running locally

• Dependencies
  • Node.js (version 20 or above)
  • Python (3.10 or above)
  • Poetry
• Clone the repository : git clone https://github.com/shubhvjain/dlonweb1.git dlonweb
• cd dlonweb
• Install all required dependencies: There are 2 possible options here.
  • Install all dependencies at once using install.sh script. Run ./install.sh from the root of the project
  • Individual components installations:
    • core : cd core && npm install
    • frontend : cd frontend && npm install
    • backend : cd backend && npm install. additionally, backend also depends on python, which m
• This can be done by installing each package individually. Or run install.sh script which will install all dependencies

• All current development happens on the main branch
• The current release version is on the release branch
• For new development, create a feature branch from main and create a PR to main
• To release a new version : create a PR from main to release