Can transformer methods be used to create fast emulators for forward and partial derivative computations in ocean modeling?
git clone [email protected]:suyashbire1/oceanfourcast.git
cd oceanfourcast
conda create --name oceanfourcast
conda activate
pip install -e .
mkdir -p data/processed/
scp [email protected]:/path/to/file/mitgcm/double_gyre/run3/dynDiag_subset.nc data/processed/. # Sample dataset
scp [email protected]:/path/to/file/mitgcm/double_gyre/run3/dynDiag.nc data/processed/. # Full dataset
python oceanfourcast/load_numpy.py --xarray_data_file "data/processed/unet/dynDiag_subset.nc" # Convert .nc to .npy
python oceanfourcast/train.py --data_file "data/processed/dynDiags.npy" --batch_size 2
# UNet
python oceanfourcast/train_unet.py --modelstr "unet" --data_file "data/processed/unet/dynDiags.npy" --batch_size 2 --output_dir "models/temp/mitgcm/unet/"