Electrostatics
python ../../train_electro.py \
--lr 5e-4 \
--ckpt_every 4000 \
--iters 32000 \
--batch_size 32 \
--train_dataset_path ../../../pde_data2/es_train_no_ma.pkl \ # if using mesh augmentation please change to es_train_ma.pkl
--test_dataset_path ../../../pde_data2/es_test.pklMagnetostatics
python ../../train_magneto.py \
--lr 5e-4 \
--ckpt_every 4000 \
--iters 32000 \
--batch_size 32 \
--train_dataset_path ../../../pde_data2/ms_train_no_ma.pkl \
--test_dataset_path ../../../pde_data2/ms_test_shape.pklFor electro, it is recomended to use following set parameter (change line 45 - 62 in tran_electro.py)
encoder = IrregSpatialEncoder2D(
input_channels=11, # the feature provided in the dataset
in_emb_dim=64,
out_chanels=64
heads=1,
depth=2,
res=50,
use_ln=False,
emb_dropout=0.05
)
decoder = IrregSpatialDecoder2D(
latent_channels=64,
out_channels=3, # the label provided in the dataset, potential and field
res=50,
scale=1,
dropout=0.0,
)For magneto, it is recomended to use following set parameter (slightly larger model)
encoder = IrregSpatialEncoder2D(
input_channels=11, # the feature provided in the dataset
in_emb_dim=64,
out_chanels=96,
heads=1,
depth=2,
res=50,
use_ln=False,
emb_dropout=0.05
)
decoder = IrregSpatialDecoder2D(
latent_channels=96,
out_channels=3, # the label provided in the dataset, potential and field
res=50,
scale=1,
dropout=0.0,
)