2019-11-30 21:32:45 +01:00
|
|
|
#!/bin/bash
|
|
|
|
|
|
|
|
|
|
#SBATCH --job-name=vel2vel
|
|
|
|
|
#SBATCH --output=%x-%j.out
|
|
|
|
|
|
|
|
|
|
#SBATCH --partition=gpu
|
|
|
|
|
#SBATCH --gres=gpu:v100-32gb:4
|
|
|
|
|
|
|
|
|
|
#SBATCH --exclusive
|
2019-12-03 23:52:01 +01:00
|
|
|
#SBATCH --nodes=4
|
2019-12-02 00:53:38 +01:00
|
|
|
#SBATCH --time=7-00:00:00
|
2019-11-30 21:32:45 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
hostname; pwd; date
|
|
|
|
|
|
|
|
|
|
|
2020-01-31 17:13:55 +01:00
|
|
|
module load gcc python3
|
|
|
|
|
#source $HOME/anaconda3/bin/activate torch
|
2019-11-30 21:32:45 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
export MASTER_ADDR=$HOSTNAME
|
2020-01-21 18:44:46 +01:00
|
|
|
export MASTER_PORT=60606
|
2019-11-30 21:32:45 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
data_root_dir="/mnt/ceph/users/yinli/Quijote"
|
|
|
|
|
|
|
|
|
|
in_dir="linear"
|
|
|
|
|
tgt_dir="nonlin"
|
|
|
|
|
|
2019-12-18 20:37:48 +01:00
|
|
|
train_dirs="*[0-8]"
|
|
|
|
|
val_dirs="*[0-8]9"
|
2019-11-30 21:32:45 +01:00
|
|
|
|
2020-01-07 02:20:05 +01:00
|
|
|
files="vel.npy"
|
2019-11-30 21:32:45 +01:00
|
|
|
in_files="$files"
|
|
|
|
|
tgt_files="$files"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
srun m2m.py train \
|
|
|
|
|
--train-in-patterns "$data_root_dir/$in_dir/$train_dirs/$in_files" \
|
|
|
|
|
--train-tgt-patterns "$data_root_dir/$tgt_dir/$train_dirs/$tgt_files" \
|
|
|
|
|
--val-in-patterns "$data_root_dir/$in_dir/$val_dirs/$in_files" \
|
|
|
|
|
--val-tgt-patterns "$data_root_dir/$tgt_dir/$val_dirs/$tgt_files" \
|
2020-01-22 00:51:32 +01:00
|
|
|
--in-norms cosmology.vel --tgt-norms cosmology.vel --augment --crop 128 --pad 20 \
|
2020-01-21 18:44:46 +01:00
|
|
|
--model VNet --adv-model UNet --cgan \
|
|
|
|
|
--lr 0.0001 --adv-lr 0.0004 --batches 1 --loader-workers 0 \
|
|
|
|
|
--epochs 128 --seed $RANDOM \
|
2019-12-18 23:06:16 +01:00
|
|
|
--cache --div-data
|
2019-12-18 20:37:48 +01:00
|
|
|
# --load-state checkpoint.pth \
|
2019-11-30 21:32:45 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
date
|
