About ASTROMER

ASTROMER's architecture is based on the Transformer architecture proposed in Vaswani (2017), with an encoder-decoder Architecture.

Pre-trained Weights

You can easily download the pre-trained weights of ASTROMER's model:

Version tag	Pretraining Data	Description	RMSE/R-Square	Link
v0	MACHO	Paper's model	0.147/0.80	Download
v1*	MACHO	Mask token and residual connections added	0.113/0.72	Download

(*) Best performance up-to-date

You can easily download the pre-trained weights of ASTROMER's model:

Description: Paper's model
Test RMSE: 0.147
Test R-Square: 0.80
Link: Download

Description: Mask token and residual connections added
Test RMSE: 0.113
Test R-Square: 0.72
Link: Download

(*) Best performance up-to-date

If you train your model from scratch, you can share your pre-trained weights by submitting a pull request in the weights repository.

You can learn more about the project in the following links:

Transformer-Based Astronomical Time Series Model with Uncertainty Estimation for Detecting Misclassified Instances (2024) . Martina Cádiz-Leyton, Guillermo Cabrera-Vives, Pavlos Protopapas, Daniel Moreno-Cartagena, Cristobal Donoso-Oliva.
Positional Encodings for Light Curve Transformers: Playing with Positions and Attention (2023). Daniel Moreno-Cartagena, Guillermo Cabrera-Vives, Pavlos Protopapas, Cristobal Donoso-Oliva, Manuel Pérez-Carrasco, Martina Cádiz-Leyton.
ASTROMER: A transformer-based embedding for the representation of light curves (2022). C. Donoso-Oliva, I. Becker, P. Protopapas, G. Cabrera-Vives, Vishnu M., Harsh Vardhan.