xgboost_chem_ages

Estimating spectroscopic stellar ages for APOGEE red-giant stars

We estimate spectroscopic stellar ages for 179 247 red-giant stars from the APOGEE DR17 catalogue (Abdurro'uf et al. 2022) with a median statistical uncertainty of 1.0 Gyr. To this end, we use the supervised machine learning technique XGBoost (Chen & Guestrin 2016), trained on a high-quality dataset of 3 060 red-giant and red-clump stars with asteroseismic ages observed by both APOGEE and Kepler (Miglio et al. 2021).

Age catalogue

• data/spec_ages_published.fits: Catalogue of spectroscopic age estimates for APOGEE DR17 stars. Duplicates are cleaned. Columns (recommended ones highlighted):

• APOGEE_ID
• spec_age_xgb: XGBoost age in Gyr (using the default model described in the paper)
• spec_age_xgb_calib: calibrated age in Gyr (adding the small correction shown in Fig. 4, top panel)
• spec_age_xgb_uncert: age uncertainty in Gyr (based on Fig. 4, bottom panel)
• spec_age_xgb_flag: Human-readable warning flag for potentially problematic stars
• spec_age_xgb_quantilereg: XGBoost quantile regression age in Gyr (using xgboost version 2.0.0)
• spec_age_xgb_quantilereg_calib: calibrated quantile regression age in Gyr
• spec_age_xgb_quantilereg_sigl: quantile regression lower 1sigma age uncertainty in Gyr
• spec_age_xgb_quantilereg_sigu: quantile regression upper 1sigma age uncertainty in Gyr
• spec_age_xgb_quantilereg_flag: Human-readable warning flag for potentially problematic stars in the quantile regression case

Jupyter notebooks

This repository contains the jupyter notebooks that allow you to reproduce the figures and analysis presented in Anders, Gispert, Ratcliffe, et al. 2023, A&A, accepted):

• train_xgboost_miglio2021.ipynb: Creating the training set, running XGBoost, and predicting ages for the APOGEE DR17 data. Reproduces Figs. 1-4 in the paper.
• train_xgboost_miglio2021_quantileregression.ipynb: Creating the training set, running XGBoost, and predicting ages for the APOGEE DR17 data. Reproduces Figs. 1-4 in the paper.
• test_age_catalogues.ipynb: Comparing the age estimates with other independent age determinations (CoRoT, K2, TESS, open clusters, [C/N] calibrations, astroNN, StarHorse, ...). Reproduces Figs. 5, 6, & A.1 in the paper.
• test_ages_science.ipynb: Testing the age estimates on some typical age-chemokinematics relations (Age-metallicity, age-[Mg/Fe] relations, radial abundance gradient as a function of age, age-velocity relation). Reproduces Figs. 7-13, A.2, A.3, B.1 in the paper.

Comments, questions, feedback are welcome: fanders[ät]icc.ub.edu