Here, we aim at deriving a cheap statistical emulator for the rheology solver in sea-ice models.
To create the environment (named ML by default):
conda env create -f environment/environment.yaml
conda activate ML
The best way to train a model is with the main.py script and a configuration file.
One example has is available out of the box (configuration in /configs/training/dv/dv-0.yaml).
cd code
python -m main --train --training_cfg dv/dv-0
That will create a results dir in /results/dv/dv-0/YYYYMMDD_HHMM/ with some quick look figures, the trained pytorch model, the dataloaders use, and a csv with validation data.