Supplementary material for: Generalizing across astronomical surveys: Few-shot light curve classification with Astromer 2

This repository contains supplementary material for the paper "Generalizing across astronomical surveys: Few-shot light curve classification with Astromer 2", accepted for publication in Astronomy & Astrophysics (2026).

To support the findings and ensure full reproducibility of the transformer-based architecture, we provide the following assets:

• Complete Manuscript Figures: A consolidated ZIP archive containing all figures presented in the paper. This includes architecture diagrams, training/validation metrics, and the detailed light curve reconstruction figures for Appendix B (EC, LPV, Cepheids, and RR Lyrae stars) that were moved here per editorial request.

• Pre-trained Foundational Weights: Dedicated files (pt_macho_v1_2021.zip and pt_macho_v2_2025.zip) containing the core encoder weights for Astromer 1 and Astromer 2, pre-trained on the MACHO survey.

• Comprehensive Model Checkpoints and Results:

  • results_v1.zip and results_v2.zip: These archives provide the complete experimental pipeline data. They include the weights of the pre-trained model, the weights of the fine-tuned models for specific downstream surveys (Alcock and ATLAS), and the weights and results of the classification stage, covering scenarios of 20, 100, and 500 samples per class.

These resources allow researchers to implement the foundational embeddings directly or build upon our few-shot classification benchmarks.