This dashboard provides a live overview of a federated training scenario. The average metric or loss curve for all nodes is provided as well as general information about the latest training run, the nodes as well as data used, and predictions of the remaining epoch and training duration. It visualises the metric and loss curves for every node as well as node connectivity states and error logging.
Just run docker compose up. This will create the three containers (server, DB, UI), start the UI on http://localhost:3000 and makes the server accessible on http://localhost:5000.
The package can be installed using:
pip install git+https://github.com/Cardio-AI/federated-dashboard.git
for example usage please refer to this example script.
The repository consists of three parts:
- Python package: This package provides an intuitive API which allows logging of events and statuses from any federated training setup to abstract the communication to the server.
- Backend server with postgres database: This part provides routes for the UI and the package.
- React client UI: This browser application visualises the data provided by the server.
This project is heavily inspired by Tensorboard and features a nearly identical API for easily integrated event logging.
It is currently in development (pull requests are welcome) but already features convenience features such as automatic updates and automatic theme detection.
