Beyond the Final Label: Exploiting the Untapped Potential of Classification Histories in Astronomical Light Curve Analysis

The Jupyter Notebook files and code for the proposed models and model training for paper named Beyond the Final Label: Exploiting the Untapped Potential of Classification Histories in Astronomical Light Curve Analysis (https://doi.org/10.48550/arXiv.2604.23792). All code and data are in the zip file.

1. lc_w_flux_*: synthetic Light curves and classsification histories for the three selected classifiers. 
2. test_data_*: held-out test sets for model evaluation.
3. lstm_atten_w_flux.ipynb: model architecture and traing for the proposed model that combines a recurrent network and attention mechanisms.
4. naive_fcn_classifier.ipynb: model architecture and training for the naive model that directly use the final classification PMFs for each object as inputs.
5. model_evaluation_demo.ipynb: model evaluation with the Early-Stable Classification Metric on baseline classifier A, with test random seed=0;
   for model evaluation on the new classifier, one need to train the model and apply the model on sequentially truncated light curves to obtain the full classification histories.
6. utils.py: helper functions for model training and data handling.