-
Notifications
You must be signed in to change notification settings - Fork 1
/
run.sh
23 lines (19 loc) · 1.57 KB
/
run.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
SEED=0
mkdir results
mkdir results/cifar10_aggre_seed_${SEED}
mkdir results/cifar10_worst_seed_${SEED}
mkdir results/cifar10_rand1_seed_${SEED}
# training
# FIRST TWO RUNS CAN RUN IN PARALLEL ON A RTX2080TI
nohup python blind_knowledge_dist_training.py --dataset cifar10 --noise_type aggre --seed ${SEED} > results/cifar10_aggre_seed_${SEED}/training.log &
nohup python blind_knowledge_dist_training.py --dataset cifar10 --noise_type worst --seed ${SEED} > results/cifar10_worst_seed_${SEED}/training.log
# THIRD RUN - Please modify with CUDA_VISIBLE_DEVICES and multiprocessing if you have the resources. Then you can run all jobs in parallel
nohup python blind_knowledge_dist_training.py --dataset cifar10 --noise_type rand1 --seed ${SEED} > results/cifar10_rand1_seed_${SEED}/training.log
# eval test_acc with learning.py
python learning.py --dataset cifar10 --noise_type aggre --seed ${SEED} > results/cifar10_aggre_seed_${SEED}/learning.log
python learning.py --dataset cifar10 --noise_type worst --seed ${SEED} > results/cifar10_worst_seed_${SEED}/learning.log
python learning.py --dataset cifar10 --noise_type rand1 --seed ${SEED} > results/cifar10_rand1_seed_${SEED}/learning.log
# eval detection-metrics with detection.py
python detection.py --dataset cifar10 --noise_type aggre --seed ${SEED} > results/cifar10_aggre_seed_${SEED}/detection.log
python detection.py --dataset cifar10 --noise_type worst --seed ${SEED} > results/cifar10_worst_seed_${SEED}/detection.log
python detection.py --dataset cifar10 --noise_type rand1 --seed ${SEED} > results/cifar10_rand1_seed_${SEED}/detection.log