Weakly-Supervised Anomaly Detection in the Milky Way 🌌

This is the companion code for https://arxiv.org/abs/2305.03761.

Quickstart

This package uses Python 3.8 and Keras 2.9.

Make sure you have conda installed on your system.

conda env create -n gaia -f requirements.yml # can also use requirements_mac.yml
conda activate gaia
python -m ipykernel install --user --name gaia --display-name "gaia"
jupyter lab

Then, navigate to one of the notebooks in the notebooks folder (making sure to specify gaia as your kernel).

Repository structure

python
├── functions.py # misc. functions, including plotting
├── models.py # define NN architecture 
├── run_mock_streams.py # apply CWoLa on 100 simulated streams
└── full_gd1_scan.py # apply CWoLa on all 21 patches covering the GD-1 stream
notebooks
├── example.ipynb # shows how to run CWoLa on a simulated stream and a real patch of GD-1
└── make_plots.ipynb # shows how to replicate each of the figures in the paper

Datasets

• Datasets needed to run notebooks/make_plots.ipynb
• Full 21 patches of GD-1

Further reading:

• CWoLa Paper
• Via Machinae 1.0
• Via Machinae 2.0
• Matt's anomaly detection workshop talk
• Gaia Dataset Info
• GD-1 Labeling