Training on Triton

[huangchieh]

Compare: with and without style translation.

How can I define two separated training processes? I need to specify where the trained POSNET model save.

I can look as this:

https://github.com/SINGROUP/ASD-AFM-dev/blob/ea21f0867353098fac6e05902406c15e0d59f9a2/asdafm/parsing_utils.py#L41

Maybe I only need to add this following parameter:

parser.add_argument("--run_dir", type=str)

Seems not work. After adding this parameter, the program also load model from Checkpoints folder.

With style transfer:

python -u 3_fit_simplified_posnet.py --train True  --test True --predict True  --epochs 2  --num_workers 1   --batch_size 4 --lr 1e-3 --avg_best_epochs 10 --pred_batches 20  --data_dir /scratch/phys/project/sin/AFM_Hartree_DB/AFM_sims/striped/Water-Au111/ --urls/train "Water-K-{1..1}_train_{0..0}.tar"  --urls/val "Water-K-{1..1}_val_{0..0}.tar"  --urls/test "Water-K-{1..1}_test_{0..0}.tar"    --peak_std 0.20    --zmin -2.5    --z_lims -2.9 0.5    --style_trans True --checkpoint_dir Checkpoints_StyleTransOn --prediction_dir ./Predictions_StyleTransOn

Without style transfer:

python -u 3_fit_simplified_posnet.py --train True  --test True --predict True  --epochs 2  --num_workers 1   --batch_size 4 --lr 1e-3 --avg_best_epochs 10 --pred_batches 20  --data_dir /scratch/phys/project/sin/AFM_Hartree_DB/AFM_sims/striped/Water-Au111/ --urls/train "Water-K-{1..1}_train_{0..0}.tar"  --urls/val "Water-K-{1..1}_val_{0..0}.tar"  --urls/test "Water-K-{1..1}_test_{0..0}.tar"    --peak_std 0.20    --zmin -2.5    --z_lims -2.9 0.5    --style_trans False --checkpoint_dir Checkpoints_StyleTransOff --prediction_dir ./Predictions_StyleTransOff