astromer1-ztfdr20

Astromer 1 (ZTF DR20 g-band)

Part of the light-curve family of open-source tools for astronomical time-series analysis.

Available from Python via the light-curve package: pip install light-curve. Documentation: light-curve.snad.space.

HuggingFace: light-curve/astromer1-ztfdr20

Paper

Nakoneczny, S. J., Bilicki, M., Pollo, A., Hui, A. Y. W., Bianco, M., Lares-Martiz, M., & Marchetti, L. (2025). QZO: A Catalog of 5 Million Quasars from the Zwicky Transient Facility. The Astrophysical Journal, 992, 153.

@article{nakoneczny2025qzo,
  author  = {Nakoneczny, S.~J. and Bilicki, M. and Pollo, A. and
             Hui, A.~Y.~W. and Bianco, M. and Lares-Martiz, M. and
             Marchetti, L.},
  title   = {{QZO}: A Catalog of 5 Million Quasars from the Zwicky
             Transient Facility},
  journal = {The Astrophysical Journal},
  volume  = {992},
  pages   = {153},
  year    = {2025},
  doi     = {10.3847/1538-4357/adcbf0},
}

Original Astromer 1 architecture:

@article{astromer1,
  author  = {Donoso-Oliva, C. and Becker, I. and Protopapas, P. and
             Cabrera-Vives, G. and Forster, F. and Est{\'e}vez, P. A.},
  title   = {{ASTROMER}: A transformer-based embedding for the representation
             of light curves},
  journal = {Astronomy \& Astrophysics},
  volume  = {670},
  pages   = {A54},
  year    = {2023},
  doi     = {10.1051/0004-6361/202243928},
}

Original code

https://github.com/snakoneczny/ztf-agn (training scripts)

Encoder architecture: https://github.com/astromer-science/main-code (v1.0 tag, git submodule at models/astromer1-ztfdr20/code/)

License

GPL-3.0 — see LICENSE.

Model overview

This is the Astromer 1 transformer encoder retrained on ZTF DR20 g-band light curves by Nakoneczny et al. (2025) for quasar/galaxy/star classification. The architecture is identical to the original MACHO-trained Astromer 1: 2 transformer layers, 4 attention heads, 256-dimensional embeddings.

The encoder weights are extracted from the ANN_clf checkpoint (full FCATT model: encoder + FC classification head) published on Zenodo. Only the encoder sublayer is exported to ONNX; the classification head is discarded.

Inputs

All tensors are float32. Both magnitudes and times are zero-mean normalized before passing to the model (subtract the per-window mean of each).

Outputs (ONNX)

Single file astromer1_ztfdr20.onnx with three named outputs:

Request only the output(s) you need via session.run(["mean"], feed) — onnxruntime will prune unused computation.

ONNX opset: 13.

Preprocessing steps

Photometric errors are not used at inference.

1. Collect ZTF g-band observation times (in days) and magnitudes.
2. Truncate each light curve to at most 200 observations.
3. Zero-mean normalize both columns over the window: time -= time.mean(), mag -= mag.mean()
4. Pad shorter light curves to exactly 200 positions: append zeros to both input and times.
5. Build the mask: set mask_in = 1 for real observations, mask_in = 0 for padded positions.
6. Reshape each tensor to [batch, 200, 1] (add trailing dimension).

The sequence length is fixed at 200 by the pretrained weights.

Weights

Source: Zenodo record 16410988 (ANN_clf.* files) Training dataset: ZTF DR20 g-band light curves cross-matched with SDSS