Finding Stellar Streams in the Milky Way with CWoLa

These datasets are designed to accompany the paper Weakly-Supervised Anomaly Detection in the Milky Way by M. Pettee, S. Thanvantri, B. Nachman, D. Shih, M. R. Buckley, and J. H. Collins (https://arxiv.org/abs/2305.03761). They illustrate how the Classification Without Labels (CWoLa) technique can be applied in the search for localized anomalies -- in this case, cold stellar streams -- in the Gaia DR2 dataset. We consider both simulated stellar streams and the known stream GD-1. The codebase is located at https://github.com/hep-lbdl/GaiaCWoLa. 

These datasets are all formatted as Pandas DataFrames, and can be loaded using `df = pd.read_hdf(file)`. Datasets ending in `*_cwola.h5*` indicate that CWoLa has been applied to the stars in that dataset. For this analysis, we consider circular "patches" of the sky of radius 15° from the Gaia DR2 dataset. 

Columns include: 

μ_δ: Proper motion (declination)

μ_α: Proper motion (right ascension)

δ: Declination

α: Right ascension

α_wrapped: Right ascension, wrapped such that all angles are > 0 

b-r: Color

g: Magnitude

ϕ: Rotated & centered right ascension 

λ: Rotated & centered declination 

μ_ϕcosλ: Rotated & centered proper motion (right ascension) 

μ_λ: Rotated & centered proper motion (declination) 

stream: Boolean label (True if part of ground truth labeling for GD-1)

patch_id: Index of the 21 patches of the GD-1 scan

nn_score: CWoLa classifier score (0 = more background-like; 1 = more signal=like)

5d_distance: For the potential GD-1 member candidates, the 5D Euclidean distance to the nearest labeled star

Files include:

gd1_1_patch.h5

Example patch of GD-1

gd1_21_patches_cwola.h5

Results from applying CWoLa to the 21 patches spanning GD-1 

promising_stars.h5

A list of stars identified by CWoLa that do not belong to the labeled GD-1 stream catalogue, but are close in 5D Euclidean space to labeled stars. Ordered in descending order by CWoLa classifier score. 

scan_bump_cwola.h5

An example of how to apply CWoLa in a scan where the stream location is unknown

scan_no_bump_cwola.h5

An example of how to apply CWoLa in a scan where the stream location is unknown

simulated_100_patches.h5

Results from applying CWoLa to 100 simulated streams 

simulated_patch.h5

A full simulated stream 

simulated_patch_cwola.h5

Results from applying CWoLa to the example simulated stream

To recreate the full scan across GD-1, one would need to use the original 21 patches of Gaia DR2 located at 10.5281/zenodo.7897935 that were constructed for arXiv:2104.12789 [astro-ph.GA). GD-1 labels were derived from 10.5281/zenodo.1295543.