First open-weight release of the morphocloud star/galaxy classifier for the DELVE-MC survey.
Model: baseline_lshsc_xgb — a calibrated XGBoost (Tier 1) classifier returning a stellar
probability P_STAR from DELVE-MC catalog photometry and morphology alone (no images, no truth
catalog at inference time). Trained with Gaia DR3, Legacy Surveys DR10 (galaxies + full-depth PSF
stars), DELVE DR3, HSC v3, and Gaia extragalactic labels; isotonic-calibrated; faint star labels
extend reliable separation to r ≈ 23 (validity floor r ≈ 23.5).
Assets (download all four into one directory):
baseline_lshsc_xgb.json— XGBoost weightsbaseline_lshsc_xgb.meta.json— feature list, params, best iterationbaseline_lshsc_xgb.calibrator.json— isotonic calibration knotsbaseline_lshsc_xgb.thresholds.csv— per-magnitude operating-point thresholds
Use: pip install morphocloud, then
StarGalaxyClassifier.load(model_path="…/baseline_lshsc_xgb.json"). Inference runs on any pandas
DataFrame / astropy Table / numpy structured array carrying the model features (build them with
features.engineer_features); cut on smooth_threshold(...) for a star/galaxy decision. See the
README and docs/model_card.md for the feature contract, evaluation, biases and the validity range.