import yaml
from scheduling import launch
def create_jobs():
jobs = [
"""python main.py --dataset imagenet --model resnet18 --opt alig
--eta 10.0 --momentum 0.0 --batch_size 1024 --epochs 90
--max_norm 400 --no_data_augmentation"""
]
return jobs
if __name__ == "__main__":
jobs = create_jobs()
launch(jobs)
import os
from scheduling import launch
jobs = [
# SGD-CE
"python train_nli.py --opt sgd --eta 1 --loss ce --no-tqdm",
# SGD-SVM
"python train_nli.py --opt sgd --eta 0.1 --loss svm --no-tqdm",
# ADAM-SVM
"python train_nli.py --opt adam --eta 1e-4 --loss svm --no-tqdm",
# ADAM-CE
"python train_nli.py --opt adam --eta 1e-4 --loss ce --no-tqdm",
# # DFW-SVM
"python train_nli.py --opt dfw --eta 1 --loss svm --no-tqdm",
]
if __name__ == "__main__":
# change current directory to InferSent
os.chdir('./InferSent/')
launch(jobs, interval=3)
# change current directory back to original
os.chdir('..')
for lr in lr_list:
jobs.append('python train.py --optimizer {optimizer} --learning_rate {lr}'
.format(optimizer=optimizer, lr=lr))
jobs.append("python train.py --optimizer alig")
def add_l4_jobs(jobs):
optimizers = ('l4adam', 'l4mom')
fraction_list = list(np.round(np.arange(0.05, 1, 0.05), 2))
for optimizer in optimizers:
for fraction in fraction_list:
jobs.append('python train.py --optimizer {optimizer} --fraction {fraction}'
.format(optimizer=optimizer, fraction=fraction))
def create_jobs():
jobs = []
add_gradient_jobs(jobs)
add_l4_jobs(jobs)
return jobs
if __name__ == "__main__":
jobs = create_jobs()
# change current directory to InferSent
os.chdir('./dnc/')
launch(jobs, on_gpu=False)
# change current directory back to original
os.chdir('..')