Strong Lensing Search: 236 Million Multiclass ML Scores

Overview: This dataset contains ~236 million multiclass machine learning classification scores generated during a search for strong gravitational lensing candidates. The scores were produced using a Vision Transformer (ViT) model trained on astronomical imaging data from the Dark Energy Survey (DES). This dataset is intended to support researchers in constructing training samples for machine learning applications to identify strong gravitational lensing systems. It provides classification scores for multiple astrophysical object categories, helping users analyze and refine candidate selections for follow-up studies. Column Definitions: COADD_OBJECT_ID: Unique object identifier from DES. RA (deg): Right Ascension (J2000). DEC (deg): Declination (J2000). SINGLE, RING, SMOOTH, COMPANIONS, SDSS_SPIRALS, DES_SPIRALS, CROWDED, ARTIFACTS, RED_SPHEROIDS: Machine learning classification scores indicating the probability that an input image belongs to a specific training class (e.g., Single, Ring, Smooth, Companions, etc.). Each score is in the range [0,1], and all class scores sum to 1 for each object. HPIX_128: HEALPix index at Nside = 128 (provides spatial information at a finer resolution). HPIX_64: HEALPix index at Nside = 64 (provides spatial information at a coarser resolution). Suggested Usage: The file is provided as a CSV file and it can be loaded using standard data analysis tools, such as Python’s Pandas library: import pandas as pddata_df = pd.read_csv('All_vit_multiclass_scores.csv')