rn50wide.sh

#!/bin/bash
#SBATCH --nodes=1
#SBATCH --gpus-per-node=1
#SBATCH --ntasks-per-node=1
#SBATCH --cpus-per-task=8
#SBATCH --job-name=cifar_supervised_rn50wide_lineareval_bn
#SBATCH --time=8:00:00
#SBATCH --array=0-47
#SBATCH --mem=64G

i=0;
for wd in 1e-6 1e-5 1e-4;
do 
	for model in resnet50w4 resnet50w2 2resnet50 4resnet50;
	do 
		for data in inaturalist18;
		do 
			wds[$i]=$wd;
			models[$i]=$model;
			datas[$i]=$data;
			i=$(($i+1));
		done 
	done
done 

final_model=${models[$SLURM_ARRAY_TASK_ID]}
final_data=${datas[$SLURM_ARRAY_TASK_ID]}
final_wd=${wds[$SLURM_ARRAY_TASK_ID]}


resdir=results/supervised/imagenet/transfer/lineareval_seer/${final_data}_${final_wd}/${final_model}/
mkdir ${resdir} -p

srun --output=${EXPERIMENT_PATH}/%j.out --error=${EXPERIMENT_PATH}/%j.err python supervised.py  --dump_path ${resdir} \
--tag supervisedimagenet_${final_model} \
--data_name ${final_data}  --classifier linear --batch_size 32 --data_path data/inaturalist18  --wd ${final_wd} \
--exp_mode lineareval --nesterov True --wd_skip_bn True \
--headinit normal  --use_bn True  --eval_freq 1  --sync_bn True  || scontrol requeue $SLURM_JOB_ID