def get_config(args):
config = get_cfg_defaults()
if args.configs:
yml_file = args.configs
config.merge_from_file(yml_file)
config.freeze()
return config
def get_config():
config = get_cfg_defaults()
if args.configs:
yml_file = args.configs
config.merge_from_file(yml_file)
config.merge_from_list(['dist_local_rank', args.local_rank])
config.freeze()
return config
def get_config(args):
config = get_cfg_defaults()
if args.configs:
yml_file = args.configs
config.merge_from_file(yml_file)
if args.csv is not None:
config.merge_from_list(['test.dataset', args.csv])
config.freeze()
return config
def get_config(configs_file):
config = get_cfg_defaults()
config.merge_from_file(configs_file)
config.freeze()
return config
step -= self.warmup_steps
lr = cosine_scheduler(step, self.base_lr, self.min_lr, self.total_steps)
return lr
def get_scheduler(config, optimizer, steps_per_epoch):
lr_scheduler = MultiScheduler(config, steps_per_epoch)
return torch.optim.lr_scheduler.LambdaLR(optimizer, lr_scheduler)
if __name__ == '__main__':
from configs.config import get_cfg_defaults
import matplotlib.pyplot as plt
config = get_cfg_defaults()
steps_per_epoches = 200
total_step = config.train.epoches * steps_per_epoches
model = torch.nn.Conv2d(3, 64, 3)
optimizer = torch.optim.Adam(model.parameters(), lr=1e-3, amsgrad=True)
# lr = MultiScheduler(config, steps_per_epoches)
# lr_scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lr)
lr_c = CosineWarmupLr(config, optimizer, steps_per_epoches)
a = []
for j in range(total_step):
optimizer.zero_grad()
# print(j, optimizer.param_groups[0]['lr'])