def test_wizard():
savedir_base = ".tmp"
hw.run_wizard(
func=test_trainval,
exp_list=[{
"lr": 1e-3
}],
savedir_base=savedir_base,
reset=0,
results_fname=f"{savedir_base}/results.ipynb",
)
shutil.rmtree(".tmp")
# 7. create main
if __name__ == '__main__':
# 8. define a list of experiments
exp_list = []
for lr in [1, 1e-1, 1e-2, 1e-3, 1e-4, 1e-5]:
exp_list += [{'lr':lr, 'dataset':'syn', 'model':'linear'}]
# 9. Launch experiments using magic command
parser = argparse.ArgumentParser()
parser.add_argument('-sb', '--savedir_base', default=None,
help='Define the base directory where the experiments will be saved.')
parser.add_argument("-r", "--reset", default=0, type=int,
help='Reset or resume the experiment.')
parser.add_argument("-j", "--run_jobs", default=0, type=int,
help='Run jobs in cluster.')
args, others = parser.parse_known_args()
if args.run_jobs == 1:
import slurm_config
job_config = slurm_config.JOB_CONFIG
elif args.run_jobs == 2:
import job_configs
job_config = job_configs.JOB_CONFIG
else:
job_config = None
hw.run_wizard(func=trainval, exp_list=exp_list, savedir_base=args.savedir_base, reset=args.reset,
job_config=job_config)
default=0,
type=int,
help='Reset or resume the experiment.')
parser.add_argument("--debug", default=False, type=int, help='Debug mode.')
parser.add_argument("-ei",
"--exp_id",
default=None,
help='Run a specific experiment based on its id.')
parser.add_argument("-j",
"--run_jobs",
default=0,
type=int,
help='Run the experiments as jobs in the cluster.')
parser.add_argument(
"-nw",
"--num_workers",
type=int,
default=0,
help='Specify the number of workers in the dataloader.')
parser.add_argument("-v",
"--visualize_notebook",
type=str,
default='',
help='Create a jupyter file to visualize the results.')
parser.add_argument("-uc", "--use_cuda", type=int, default=1)
args, others = parser.parse_known_args()
# 9. Launch experiments using magic command
hw.run_wizard(func=trainval, exp_groups=exp_configs.EXP_GROUPS, args=args)
model = he.get_model(name=exp_dict['model'], exp_dict=exp_dict)
# 3. load checkpoint
chk_dict = hw.get_checkpoint(savedir)
# 4. Add main loop
for epoch in tqdm.tqdm(range(chk_dict['epoch'], 10),
desc="Running Experiment"):
# 5. train for one epoch
train_dict = model.train_on_loader(train_loader, epoch=epoch)
# 6. get and save metrics
score_dict = {'epoch':epoch, 'acc': train_dict['train_acc'],
'loss':train_dict['train_loss']}
chk_dict['score_list'] += [score_dict]
images = model.vis_on_loader(train_loader)
hw.save_checkpoint(savedir, score_list=chk_dict['score_list'], images=[images])
print('Experiment done\n')
# 7. create main
if __name__ == '__main__':
# 8. define a list of experiments
exp_list = []
for lr in [1, 1e-1, 1e-2, 1e-3, 1e-4, 1e-5, "bug"]:
exp_list += [{'lr':lr, 'dataset':'mnist', 'model':'linear'}]
# 9. Launch experiments using magic command
hw.run_wizard(func=trainval, exp_list=exp_list, job_config=job_config.JOB_CONFIG)
# Savedir Arguments
parser.add_argument('-sb', '--savedir_base', required=True)
parser.add_argument('-d', '--datadir', required=True)
# Others
parser.add_argument("-r", "--reset", default=0, type=int)
parser.add_argument("-j", "--run_jobs", default=0, type=int)
parser.add_argument("-v", "--visualize_notebook", type=str, default='')
args, others = parser.parse_known_args()
# -- Launch Experiments
# Get Experiment Groups
import exp_configs
exp_groups = exp_configs.EXP_GROUPS
print('Launching exp_group: %s' % args.exp_group_list)
if os.path.exists('job_configs.py'):
import job_configs
job_config = job_configs.JOB_CONFIG
else:
job_config = None
# Run Selected Experiments
hw.run_wizard(func=trainval,
exp_groups=exp_groups,
args=args,
job_config=job_config)
args, others = parser.parse_known_args()
# Define a list of experiments
if args.exp_group == "syn":
exp_list = []
for lr in [1, 1e-1, 1e-2, 1e-3, 1e-4, 1e-5]:
exp_list += [{"lr": lr, "dataset": "syn", "model": "linear"}]
# Choose Job Scheduler
if args.job_scheduler == "slurm":
import slurm_config
job_config = slurm_config.JOB_CONFIG
elif args.job_scheduler == "toolkit":
import job_configs
job_config = job_configs.JOB_CONFIG
else:
job_config = None
# Run experiments and create results file
hw.run_wizard(
func=trainval,
exp_list=exp_list,
savedir_base=args.savedir_base,
reset=args.reset,
job_config=job_config,
results_fname="results.ipynb",
)
score_dict = {
'epoch': epoch,
'acc': train_dict['train_acc'],
'loss': train_dict['train_loss']
}
chk_dict['score_list'] += [score_dict]
images = model.vis_on_loader(train_loader)
hw.save_checkpoint(savedir,
score_list=chk_dict['score_list'],
images=[images])
print('Experiment done\n')
# 7. create main
if __name__ == '__main__':
# 8. define a list of experiments
exp_list = []
for lr in [1, 1e-1, 1e-2, 1e-3, 1e-4, 1e-5, "bug"]:
exp_list += [{'lr': lr, 'dataset': 'mnist', 'model': 'linear'}]
# 9. Launch experiments using magic command
if os.path.exists('job_config.py'):
import job_config
jc = job_config.JOB_CONFIG
else:
jc = None
hw.run_wizard(func=trainval, exp_list=exp_list, job_config=jc)
ut.torch_save(model_path, model.get_state_dict())
print("Saved: %s" % savedir)
if __name__ == "__main__":
import exp_configs
parser = argparse.ArgumentParser()
parser.add_argument('-e', '--exp_group_list', nargs="+")
parser.add_argument('-sb', '--savedir_base', required=True)
parser.add_argument('-d', '--datadir', required=True)
parser.add_argument("-r", "--reset", default=0, type=int)
parser.add_argument("-c", "--cuda", default=1, type=int)
parser.add_argument("-j", "--job_scheduler", default=None)
parser.add_argument("-p", "--python_binary_path", default=None)
args, others = parser.parse_known_args()
# Get job configuration to launch experiments in the cluster
job_config = None
if os.path.exists('job_configs.py'):
import job_configs
job_config = job_configs.JOB_CONFIG
# Run experiments either sequentially or in the cluster
hw.run_wizard(func=trainval,
exp_groups=exp_configs.EXP_GROUPS,
job_config=job_config,
job_scheduler=args.job_scheduler,
python_binary_path=args.python_binary_path,
use_threads=True,
args=args)
