Add wide-buffer dataset for training-time pick-jitter augmentation. Same trace survival/order as processed/ (same exclusion rules, evaluated against the nominal window) with a real-signal margin captured on each side -- see seismic_ensemble.preprocessing.apply_wide.
Add wide-buffer dataset for training-time pick-jitter augmentation. Same trace survival/order as processed/ (same exclusion rules, evaluated against the nominal window) with a real-signal margin captured on each side -- see seismic_ensemble.preprocessing.apply_wide.
all checkpoints and experimental outcomes for the paper titled: Robust and Explainable Multi-Class Seismic Event Discrimination through Benchmarking and Ensemble Learning (part 4)
all checkpoints and experimental outcomes for the paper titled: Robust and Explainable Multi-Class Seismic Event Discrimination through Benchmarking and Ensemble Learning (part 3)
all checkpoints and experimental outcomes for the paper titled: Robust and Explainable Multi-Class Seismic Event Discrimination through Benchmarking and Ensemble Learning (part 2)
all checkpoints and experimental outcomes for the paper titled: Robust and Explainable Multi-Class Seismic Event Discrimination through Benchmarking and Ensemble Learning
Fix noise-class padding shortcut: no_onset_start_s 10.0 -> 0.0. Every STEAD noise trace previously padded an identical, deterministic 200 samples (10s), a zero-exception class signature every model could trivially key on instead of genuine content -- see scripts/check_noise_padding_shortcut.py.
Fix noise-class padding shortcut: no_onset_start_s 10.0 -> 0.0. Every STEAD noise trace previously padded an identical, deterministic 200 samples (10s), a zero-exception class signature every model could trivially key on instead of genuine content -- see scripts/check_noise_padding_shortcut.py.
Fix noise-class padding shortcut: no_onset_start_s 10.0 -> 0.0. Every STEAD noise trace previously padded an identical, deterministic 200 samples (10s), a zero-exception class signature every model could trivially key on instead of genuine content -- see scripts/check_noise_padding_shortcut.py.
