DES_DR2_star_galaxy_samples_updated

This Zenodo record provides a complete, citable, and reproducible package for the experiments and figures in the manuscript: “Reproducible star–galaxy separation in DES DR2 18 ≤ i < 24: a minimal machine-learning baseline with slice-wise metrics, calibration diagnostics, and visualization mosaics”(Submitted to Astronomy & Computing; manuscript ID: ASCOM-D-25-00224R1) Overview The package contains: Magnitude-stratified DES DR2 samples used for visualization and baseline experiments, All manuscript figures and their source plotting scripts, All experiment output tables (slice-wise metrics, class balance, calibration diagnostics, cross-validation comparisons, seeing-stratified results, augmentation deltas), Both legacy exploratory sampling code and the final sampling workflow, and The full reproducible analysis notebook(s) used to regenerate the main results. The primary goal is to enable readers to exactly reproduce the sampling, figures, and tables reported in the manuscript, and to provide transparent reference outputs for comparison. Contents A. Data products (samples) Stratified samples from DES DR2 covering the i-band magnitude range 18 ≤ MAG_AUTO_I <24. Each sample file contains randomly selected objects with (at minimum) the columns: mag_auto_i spread_model_i extended_class_coadd(Additional columns may be present in derived tables depending on experiment stage.) Sampling scheme (legacy workflow): 18 ≤ i <19.25 : 10,000 objects per 0.25-mag bin 19.25 ≤ i < 19.5: 5,000 objects per 0.125-mag bin 19.50 ≤ i < 19.75: 7,500 objects per 0.25-mag bin 19.75 ≤ i < 24: 5,000 objects per 0.125-mag bin Sampling scheme (final workflow): 18 ≤ i < 20 : 83,333 objects per 0.5-mag bin 20 ≤ i < 22: 83,334 objects per 0.5-mag bin 22 ≤ i < 24: 83,333 objects per 0.5-mag bin Samples are distributed as plain-text / CSV-style tables (tab- or comma-separated as indicated in file headers). A compressed archive contains all per-bin/per-slice sample files. B. Reproducible code (sampling + analysis) This record includes three main notebooks (and any helper scripts referenced by them): Legacy exploratory sampling SQL notebook: early/diagnostic sampling queries and checks. Final magnitude-stratified sampling SQL notebook: the finalized query logic used to produce the stratified sample products. ASCOM R2 analysis notebook: end-to-end analysis producing slice-wise logistic-regression results, metrics, calibration diagnostics, seeing experiments, and augmentation tests. All notebooks are provided as runnable .ipynb files. They are designed to regenerate the released plots and tables using the included intermediate products (or by rerunning SQL sampling when the relevant DES DR2 access is available to the user). C. Pre-generated manuscript outputs To support immediate inspection and citation, the record also includes: All manuscript figures as rendered image files (e.g., PNG/PDF), including: global data-space visualization (MAG_AUTO_I vs SPREAD_MODEL_I), slice-wise performance curves (precision/recall/F1/AUC), calibration diagnostics (Brier score, skill score, reliability curves), seeing-stratified analyses, augmentation deltas and cross-validation comparisons, mosaic visualization sets (legacy and revised). All experiment tables as machine-readable files (CSV), including (naming may vary by version): per-slice class balance / extended fraction π, metrics vs slice for random CV and tile-wise CV, block-minus-random deltas, augmentation deltas, seeing quartiles and seeing-mapped performance summaries. These pre-generated outputs allow readers to verify the manuscript results without running any code. Reproducibility notes The notebooks are intended to be run in a standard Python scientific environment (NumPy/Pandas/Matplotlib/scikit-learn). Exact regeneration of DES DR2 queries requires appropriate data access; however, the released intermediate sample products and derived tables are included so that all manuscript plots and metrics can be reproduced from the Zenodo package alone. Provenance This dataset and codebase were developed as part of ongoing reproducible-methods work at The Ohio State University and directly support the above manuscript submission to Astronomy & Computing. Earlier exploratory sampling materials are included for transparency, while the finalized R2 workflow represents the version used for the resubmission. License and citation Please cite this Zenodo record using its DOI in any academic work that uses these samples, tables, figures, or code. If you reuse substantial parts of the workflow, please also cite the associated manuscript (once published).