[Phase 5] Cross-Survey Validation

[Sakeeb91]

Objective

Validate models on GALAH overlap sample and assess systematic differences between surveys.

Dependencies

• Phase 4: Parameter Regression (requires trained models)

Tasks

• Implement src/data/galah_loader.py for GALAH data parsing
• Implement src/data/crossmatch.py for coordinate cross-matching
• Cross-match APOGEE and GALAH by sky coordinates
• Compare APOGEE predictions to GALAH labels
• Quantify systematic offsets between surveys
• Apply calibration if needed
• Create notebooks/04_cross_validation.ipynb

Files to Create

File	Purpose
src/data/galah_loader.py	Load GALAH data
src/data/crossmatch.py	Cross-match utilities
src/evaluation/cross_survey.py	Cross-survey metrics

Starter Code

# src/data/crossmatch.py
"""Cross-match utilities for multi-survey data."""

import numpy as np
from astropy.coordinates import SkyCoord
from astropy import units as u

def crossmatch_by_coordinates(
    ra1: np.ndarray, dec1: np.ndarray,
    ra2: np.ndarray, dec2: np.ndarray,
    max_sep_arcsec: float = 2.0
) -> tuple[np.ndarray, np.ndarray, np.ndarray]:
    """
    Cross-match two catalogs by sky coordinates.

    Returns indices of matches and separations.
    """
    coords1 = SkyCoord(ra=ra1*u.deg, dec=dec1*u.deg)
    coords2 = SkyCoord(ra=ra2*u.deg, dec=dec2*u.deg)

    idx, sep2d, _ = coords1.match_to_catalog_sky(coords2)

    # Filter by maximum separation
    mask = sep2d.arcsec < max_sep_arcsec

    idx1 = np.where(mask)[0]
    idx2 = idx[mask]
    separations = sep2d[mask].arcsec

    return idx1, idx2, separations

Definition of Done

• Cross-match identifies >10,000 common stars
• Systematic offsets quantified (Teff < 50K median difference)
• Calibration applied if offsets significant
• Results documented in notebook
• All tests passing

Technical Notes

• Use 2 arcsec matching radius for cross-match
• Different wavelength coverage may cause systematic differences
• Document any calibration choices for reproducibility

---

Part of #1 (Meta Issue)