poly學習政策
當power<0,呈現為凸函數,随着epoch增加,學習率下降速率由慢變快
當power >1,呈現為凹函數,随着epoch增加,學習率下降速率由快變慢
def adjust_lr_poly(optimizer, epoch, num_epochs, base_lr, power=0.9)
lr = base_lr * (1-epoch/num_epochs)**power
for param_group in optimizer.param_groups:
param_group['lr'] = lr
return lr