paper : Bag of Tricks for Image Classification with Convolutional Neural Networks (https://arxiv.org/abs/1812.01187)
python scheduler.py
from warmup_scheduler.scheduler import GradualWarmupScheduler
v = torch.zeros(10)
optim = torch.optim.SGD([v], lr=0.01)
cosine_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optim, 100, eta_min=0, last_epoch=-1)
scheduler = GradualWarmupScheduler(optim, multiplier=8, total_epoch=5, after_scheduler=cosine_scheduler)
for epoch in range(1, 100):
scheduler.step(epoch)
max_epoch = num
for epoch in range(1, max_epoch):
cosine_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optim, max_epoch, eta_min=0, last_epoch=-1)
To change the epoch, change all of the highlighted text.