YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Sparse light-curve early classification — trained weights
Best checkpoints for A Comparative Study of Machine Learning Methods for Early Classification from Sparse Astronomical Light Curves.
License: Apache License 2.0
Code: Yu-Yang-Li/StarWhisper → directory Early Classification from Sparse Light Curves
This repository contains best checkpoints only (best_model.pth / *_best.json), plus small companion files (scaler.pkl, label_encoder.pkl, feature_columns.pkl, best_params.txt where applicable).
Contents
| Directory | File(s) | Approx. size | Role |
|---|---|---|---|
xgboost_full/ |
xgboost_full_best.json |
~490 MB | XGBoost, 57-feature set |
xgboost_reduced/ |
xgboost_reduced_best.json |
~408 MB | XGBoost, 39-feature set |
transformer_feat_50obs_pretrain/ |
best_model.pth + scaler/encoder |
~868 MB | Feature Transformer, 50-obs pretrain |
transformer_feat_varlen_finetune/ |
best_model.pth + scaler/encoder |
~868 MB | Feature Transformer, varlen finetune |
transformer_feat_varlen_scratch/ |
best_model.pth + scaler/encoder |
~868 MB | Feature Transformer, varlen scratch |
e2e_transformer_50obs_pretrain/ |
best_model.pth |
~41 MB | E2E Transformer (lightweight), 50-obs |
e2e_transformer_varlen_finetune/ |
best_model.pth |
~41 MB | E2E Transformer (lightweight), finetune |
e2e_transformer_50obs_matched_pretrain/ |
best_model.pth |
~868 MB | E2E Transformer (matched size), 50-obs |
e2e_transformer_varlen_matched_finetune/ |
best_model.pth |
~868 MB | E2E Transformer (matched size), finetune |
e2e_transformer_varlen_matched_scratch/ |
best_model.pth |
~868 MB | E2E Transformer (matched size), scratch |
lstm_varlen/ |
best_model.pth |
~0.8 MB | LSTM E2E, varlen |
Directory names match train_models/ in the code repository. Load and evaluate with the corresponding training/evaluation scripts; experiment IDs and metric paths are listed in train_models/experiment_registry.py.
Citation
Please cite the associated paper when using these weights.