Loss Lens

LossLens is a comprehensive visual analytics framework provides an interactive, multi-scale visual representation and analytical pipeline for exploring loss landscapes of machine learning models.

Run via Docker

To run the system through Docker, run the following under the root directory:

docker-compose up

Then, you should be able to launch the system via localhost:3000.

Run Locally

Running locally needs a bit more step. Before installing the LossLens, make sure you have Nodejs, Python, and MongoDB installed first. Version requirements:

Software	Min. Version
Nodejs	18.0
Python	3.8
MongoDB	6.0

Install GUI Dependencies

Under the gui folder, run

npm install

to install all dependencies.

Install Back-end Dependencies

Under the api folder, run

pip install -r requirements.txt
pip install git+https://github.com/fra31/auto-attack
pip install git+https://github.com/RobustBench/robustbench.git

to install all dependencies.

Restore the Data on Local Machine

Under the database folder, run

mongorestore --host mongodb --port 27017 --db losslensdb ./losslensdb/

Run LossLens

Run

npm run dev

to start the frontend server.

Run

python server.py

to start the backend server.

Adding case study

please see adding case study

Contributing

We are welcoming contributions from our collaborators. We strongly encourage contributors to work in their own forks of this project. You can edit the code on your own branch after you create your own forks. When you are ready to contribute, you can click the Compare & pull request button next to your own branch under your repository.

This project highly recommand committers to sign their code using the Developer Certificate of Origin (DCO) for each git commit but this is not required.