FlareSense v2

FlareSense is an experimental project that trains a convolutional neural network on e-Callisto radio spectrograms to detect solar radio bursts. The repository contains training code, prediction utilities, a web app for user interaction, and several notebooks used during development.

Installation

The code requires Python 3.11. Use the conda environment for dependencies:

conda activate flaresense-v2
pip install -r requirements.txt

Training

Training is driven by YAML configuration files. A typical run looks like:

python main.py --config configs/best_v2.yml

The main.sh script shows how to submit a job on a SLURM cluster.

Prediction

Run inference on a dataset with:

python pred_dataset.py

For live prediction through a Gradio interface execute:

python pred_live.py

Web App

The web app in app.py provides a user interface for uploading data, running predictions, and viewing results. It is built with Flask and integrates with the prediction models.

Evaluation

To reproduce our results, run the following command:

python main.py --config configs/best_v2.yml

Notebooks

All notebooks can be found in the _notebooks directory. They provide exploratory data analysis, model investigations, and visualizations.

Deployment

FlareSense is deployed as Linux systemd services for production use:

• flaresense_app.service: Manages the web app (app.py) for user-facing interactions.
• flaresense.service: Handles continuous prediction tasks.

Service Management

If FlareSense is deployed as systemd services, you can inspect the logs with:

sudo journalctl -u flaresense_app.service  # For web app logs
sudo journalctl -u flaresense.service      # For prediction service logs

After modifying the service files, redeploy with:

sudo systemctl restart flaresense_app.service
sudo systemctl restart flaresense.service
sudo systemctl daemon-reload

The .serice-file can be found here: /etc/systemd/system/flaresense_app.service

This repository is provided for reference without any warranty.