ocean-forecast-training

What problem does it solve?

Ocean scientists and ML engineers need reliable, repeatable workflows to train ocean forecast models on large time-series datasets, with safe defaults and robust error handling.

Core Features & Use Cases

• End-to-end orchestration: dataset validation, model selection, training execution, and result reporting.
• Model-agnostic training: supports FNO2d, UNet2d, SwinTransformerV2, Crossformer, and other Forecast models with guided hyperparameters and safe guards.
• Operational safety rails: OOM protection, automatic parameter suggestions, and failure recovery workflows.

Quick Start

Launch an end-to-end ocean forecast training workflow by supplying dataset_root and log_dir to ocean_forecast_train_start and then confirming the model and parameters to begin training.