Datasets

Training datasets

Labeled datasets to train algorithms, provided to get comfortable with the format and processing. Targets are included for training, but the full parameter space in the final datasets is incomplete. The goal is to learn the behavior and prepare for extrapolation.

Validation/evaluation datasets

A private validation dataset lets participants test models before final submission. Evaluation metrics match the final scoring. Validation and inference run on the Codabench server (A10s on SDSC’s NSF National Research Platform).

Final Challenge datasets

Performance on the final challenge datasets determines scores. Each sub-challenge provides scoring code in its GitHub repository. After the competition, these datasets will be public. Scoring is performed on Codabench (A10s on SDSC’s NSF National Research Platform).