Skip to content

MODEL_CARD.md

Latest commit

 

History

History
57 lines (38 loc) · 1.42 KB

MODEL_CARD.md

File metadata and controls

57 lines (38 loc) · 1.42 KB
Raw

PISCES model card

Artifact

  • Model: model/pisces_preview.pt
  • Runner: pisces_preview/run_preview.py

Input

Normalized 60-step windows with seven channels:

bx, by, bz, bt, density, speed, temperature

Optional start_utc and source metadata columns are allowed.

Output

window_id,reconstruction_error

If start_utc or source is present in the input, it is copied to the output. reconstruction_error is unweighted mean squared reconstruction error in normalized input space.

Examples

Synthetic example:

python -m pisces_preview.run_preview examples/example_normalized_windows.csv --out preview_output.csv

NASA OMNI preview windows:

python -m pisces_preview.run_preview examples/omni_may2024_preview_windows.csv --out omni_preview_output.csv

Expected OMNI output is in examples/expected_preview_output.csv.

Scope

This preview includes the initial model, small normalized inputs, and supporting materials. Full preprocessing, scoring, thresholds, event evaluation, training, and baselines are planned for the full release.

Citation

@inproceedings{march2026pisces,
  title={{PISCES}: Physics-Informed Convolutional Autoencoder for Solar Wind Anomaly
         Detection and Space Weather Early Warning},
  author={March, Alison J. and Lee, Kevin},
  booktitle={NASA 5th Eddy Cross-Disciplinary Symposium},
  year={2026},
  address={Boulder, Colorado}
}